Optimized Exit bridge managment

This adds commits 6ff68849f88cbec622fe7a485e301111319ab8d6 and
53687387ebca091043cecbfc2d99494865336b94 from box64.

This allows installation of box86 without any configuration files or
libraries involved. Intended usecases are

 * packaging box86 with multiple variants in one package, where multiple
   variants would step on each other during installation.
 * packaging box86 for distributions shipping their own armhf shared
   library versions of libc and c++ (for example via multiarch)

.github/workflows/release.yml

@@ -34,21 +34,21 @@ jobs:
         include:
           - platform: TEGRAX1
             type: Release
-            os: ubuntu-18.04
+            os: ubuntu-20.04
           - platform: TEGRAX1
             type: Trace
-            os: ubuntu-18.04
+            os: ubuntu-20.04
           - platform: TEGRAX1ARM32
             type: Release
-            os: ubuntu-18.04
+            os: ubuntu-20.04
           - platform: TEGRAX1ARM32
             type: Trace
-            os: ubuntu-18.04
+            os: ubuntu-20.04
 
     runs-on: ${{ matrix.os }}
     steps:
       - name: "Checkout Box86 Repository"
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
 
       - name: "Environment preparation"
         run: |
@@ -111,7 +111,7 @@ jobs:
           ctest -j$(nproc)
 
       - name: "Upload Artifact"
-        uses: actions/upload-artifact@v2
+        uses: actions/upload-artifact@v3
         with:
           name: box86-${{ matrix.platform }}-${{ matrix.type }}
           path: build/box86

.gitignore

@@ -5,4 +5,5 @@ build*/
 src/git_head.h
 backup/
 box86_update.sh
-restore_backup.sh
+restore_backup.sh
+.cache

CMakeLists.txt

@@ -5,10 +5,12 @@ option(PYRA "Set to ON if targeting an OpenPandora device" ${PYRA})
 option(RPI2 "Set to ON if targeting an RaspberryPI2 device" ${RPI2})
 option(RPI3 "Set to ON if targeting an RaspberryPI3 device" ${RPI3})
 option(RPI4 "Set to ON if targeting an RaspberryPI4 device" ${RPI4})
+option(RPI3ARM64 "Set to ON if targeting an RaspberryPI3 device with multiarch arm64 and armhf" ${RPI3ARM64})
 option(RPI4ARM64 "Set to ON if targeting an RaspberryPI4 device with multiarch arm64 and armhf" ${RPI4ARM64})
 option(GOA_CLONE "Set to ON if targeting GO Advance clones, like RG351p/v, Gameforce Chi, RGB10..." ${GOA_CLONE})
 option(RK3288 "Set to ON if targeting an Rockchip RK3288 based device" ${RK3288})
 option(RK3399 "Set to ON if targeting an Rockchip RK3399 based device" ${RK3399})
+option(RK3588 "Set to ON if targeting an Rockchip RK3588 / RK3588S based device" ${RK3588})
 option(ODROIDN2 "Set to ON if targeting an Odroid-N2 device" ${ODROIDN2})
 option(TEGRAX1 "Set to ON if targeting an Tegra X1 (ARM64) based device" ${TEGRAX1})
 option(TEGRAX1ARM32 "Set to ON if targeting an Tegra X1 (ARM32) based device" ${TEGRAX1ARM32})
@@ -18,6 +20,8 @@ option(ODROID "Set to ON if targeting an ODroid device" ${ODROID})
 option(POWER9 "Set to ON if targeting a POWER9 processor" ${POWER9})
 option(SD845 "Set to ON if targeting a Snapragon 845 based device" ${SD845})
 option(A64 "Set to ON if targeting an Allwinner A64 based device" ${A64})
+option(ARM64 "Set to ON if targeting a generic ARM64 based device" ${ARM64})
+option(BAD_SIGNAL "Set to ON to activate the workaround for incoherent si_info on SIGSEGV" ${BAD_SIGNAL})
 option(LX2160A "Set to ON if targeting an LX2160A based device" ${LX2160A})
 option(ANDROID "Set to ON if targeting an Android device" ${ANDROID})
 option(USE_CCACHE "Set to ON to use ccache if present in the system" ${USE_CCACHE})
@@ -25,10 +29,16 @@ option(HAVE_TRACE "Set to ON to have Trace ability (needs ZydisInfo library)" ${
 option(NOLOADADDR "Set to ON to avoid fixing the load address of Box86" ${NO_LOADAADR})
 option(NOGETCLOCK "Set to ON to avoid using clock_gettime with CLOCK_MONOTONIC_COARSE for RDTSC opcode (use gltimeofday instead)" ${NOGETCLOCK})
 option(NOGIT "Set to ON if not building from a git clone repo (like when building from a zip download from github)" ${NOGIT})
-if(PANDORA OR PYRA OR RPI2 OR RPI3 OR RPI4 OR GAMESHELL OR ODROID OR GOA_CLONE OR RK3288 OR RK3399 OR ODROIDN2 OR TEGRAX1 OR TEGRAX1ARM32 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A)
+option(NO_LIB_INSTALL "Set ON to not install a few x86 libs that are used by many program" ${NO_LIB_INSTALL})
+option(NO_CONF_INSTALL "Set ON to not install config files" ${NO_CONF_INSTALL})
+
+if(PANDORA OR PYRA OR RPI2 OR RPI3 OR RPI4 OR GAMESHELL OR ODROID OR GOA_CLONE OR RK3288 OR RK3399 OR ODROIDN2 OR TEGRAX1 OR TEGRAX1ARM32 OR RPI3ARM64 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A OR ARM64 OR RK3588)
     set(LD80BITS OFF CACHE BOOL "")
     set(NOALIGN OFF CACHE BOOL "")
     set(ARM_DYNAREC ON CACHE BOOL "")
+    if(RK3588)
+        set(BAD_SIGNAL ON CACHE BOOL "")
+    endif()
 endif()
 if(ANDROID)
     set(NOLOADADDR ON CACHE BOOL "")
@@ -61,7 +71,7 @@ string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "x86_64"  _x86_64)
 string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "aarch64"  _aarch64)
 string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "armhf"  _armhf)
 
-if(_aarch64 AND (RK3399 OR ODROIDN2 OR TEGRAX1 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A))
+if(_aarch64 AND (RK3399 OR ODROIDN2 OR TEGRAX1 OR RPI3ARM64 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A OR RK3588 OR ARM64))
     set(CMAKE_C_COMPILER arm-linux-gnueabihf-gcc)
     #set(CMAKE_ASM_COMPILER arm-linux-gnueabihf-as) #will use gcc in fact
     set(CMAKE_C_COMPILER_TARGET arm-linux-gnueabihf)
@@ -81,6 +91,7 @@ endif()
 if(PANDORA)
     add_definitions(-DPANDORA)
     #add_definitions(-mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp -ftree-vectorize -fsingle-precision-constant -ffast-math)
+    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp")
 elseif(GAMESHELL)
     add_definitions(-DGAMESHELL)
     add_definitions(-marm -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard)
@@ -104,6 +115,11 @@ elseif(RPI4)
     add_definitions(-DRPI4)
     add_definitions(-marm -march=armv8-a+crc -mtune=cortex-a72 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
     set(CMAKE_ASM_FLAGS  "-marm -march=armv8-a+crc -mtune=cortex-a72 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(RPI3ARM64)
+    add_definitions(-DRPI)
+    add_definitions(-DRPI3ARM64)
+    add_definitions(-marm -march=armv8-a+crc -mtune=cortex-a53 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -march=armv8-a+crc -mtune=cortex-a53 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
 elseif(RPI4ARM64)
     add_definitions(-DRPI)
     add_definitions(-DRPI4ARM64)
@@ -133,6 +149,15 @@ elseif(RK3399)
     add_definitions(-DRK3399)
     add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a72+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
     set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a72+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(RK3588)
+    add_definitions(-DRK3588)
+    add_definitions(-pipe -mcpu=cortex-a76.cortex-a55+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    add_definitions(-DBAD_SIGNAL)
+    set(CMAKE_ASM_FLAGS  "-pipe -mcpu=cortex-a76.cortex-a55+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(ARM64)
+    add_definitions(-DARM64)
+    add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
 elseif(ODROIDN2)
     add_definitions(-DODROIDN2)
     add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a73.cortex-a53+crypto -mfpu=crypto-neon-fp-armv8 -mfloat-abi=hard)
@@ -160,6 +185,10 @@ elseif(ARM_DYNAREC)
     set(CMAKE_ASM_FLAGS  "-pipe -march=armv7-a+simd -mfpu=neon")
 endif()
 
+if(BAD_SIGNAL)
+    add_definitions(-DBAD_SIGNAL)
+endif()
+
 if(POWER9)
   add_definitions(-mcpu=power9 -mtune=power9)
 endif()
@@ -260,6 +289,7 @@ set(ELFLOADER_SRC
     "${BOX86_ROOT}/src/tools/box86stack.c"
     "${BOX86_ROOT}/src/tools/my_cpuid.c"
     "${BOX86_ROOT}/src/tools/gtkclass.c"
+    "${BOX86_ROOT}/src/tools/rcfile.c"
     "${BOX86_ROOT}/src/tools/wine_tools.c"
     "${BOX86_ROOT}/src/elfs/elfloader.c"
     "${BOX86_ROOT}/src/elfs/elfparser.c"
@@ -270,9 +300,6 @@ set(ELFLOADER_SRC
     "${BOX86_ROOT}/src/librarian/symbols.c"
     "${BOX86_ROOT}/src/librarian/globalsymbols.c"
     "${BOX86_ROOT}/src/emu/x86emu.c"
-    "${BOX86_ROOT}/src/emu/x86run.c"
-    "${BOX86_ROOT}/src/emu/x86run66.c"
-    "${BOX86_ROOT}/src/emu/x86run660f.c"
     "${BOX86_ROOT}/src/emu/x86run_private.c"
     "${BOX86_ROOT}/src/emu/x86syscall.c"
     "${BOX86_ROOT}/src/emu/x86primop.c"
@@ -280,7 +307,6 @@ set(ELFLOADER_SRC
     "${BOX86_ROOT}/src/emu/x86int3.c"
     "${BOX86_ROOT}/src/emu/x86tls.c"
     "${BOX86_ROOT}/src/emu/x87emu_private.c"
-    "${BOX86_ROOT}/src/emu/x87run.c"
     "${BOX86_ROOT}/src/libtools/auxval.c"
     "${BOX86_ROOT}/src/libtools/myalign.c"
     "${BOX86_ROOT}/src/libtools/myalign64.c"
@@ -297,6 +323,33 @@ if(NOT ANDROID)
     list(APPEND ELFLOADER_SRC "${BOX86_ROOT}/src/libtools/obstack.c")
 endif()
 
+set(INTERPRETER
+    "${BOX86_ROOT}/src/emu/x86run.c"
+    "${BOX86_ROOT}/src/emu/x86run0f.c"
+    "${BOX86_ROOT}/src/emu/x86run64.c"
+    "${BOX86_ROOT}/src/emu/x86run640f.c"
+    "${BOX86_ROOT}/src/emu/x86run6466.c"
+    "${BOX86_ROOT}/src/emu/x86run6467.c"
+    "${BOX86_ROOT}/src/emu/x86run66.c"
+    "${BOX86_ROOT}/src/emu/x86run660f.c"
+    "${BOX86_ROOT}/src/emu/x86run66d9.c"
+    "${BOX86_ROOT}/src/emu/x86run66dd.c"
+    "${BOX86_ROOT}/src/emu/x86run67.c"
+    "${BOX86_ROOT}/src/emu/x86run6766.c"
+    "${BOX86_ROOT}/src/emu/x86rund8.c"
+    "${BOX86_ROOT}/src/emu/x86rund9.c"
+    "${BOX86_ROOT}/src/emu/x86runda.c"
+    "${BOX86_ROOT}/src/emu/x86rundb.c"
+    "${BOX86_ROOT}/src/emu/x86rundc.c"
+    "${BOX86_ROOT}/src/emu/x86rundd.c"
+    "${BOX86_ROOT}/src/emu/x86runde.c"
+    "${BOX86_ROOT}/src/emu/x86rundf.c"
+    "${BOX86_ROOT}/src/emu/x86runf0.c"
+    "${BOX86_ROOT}/src/emu/x86runf066.c"
+    "${BOX86_ROOT}/src/emu/x86runf20f.c"
+    "${BOX86_ROOT}/src/emu/x86runf30f.c"
+)
+
 set(WRAPPEDS
     "${BOX86_ROOT}/src/wrapped/wrappedlibc.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibpthread.c"
@@ -327,6 +380,7 @@ set(WRAPPEDS
     "${BOX86_ROOT}/src/wrapped/wrappedpng16.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxxf86vm.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxrandr.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxpresent.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxext.c"
     "${BOX86_ROOT}/src/wrapped/wrappedxinerama.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxfixes.c"
@@ -408,6 +462,7 @@ set(WRAPPEDS
     "${BOX86_ROOT}/src/wrapped/wrappedgsttag.c"
     "${BOX86_ROOT}/src/wrapped/wrappedgstbase.c"
     "${BOX86_ROOT}/src/wrapped/wrappedgstapp.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibibus.c"
     "${BOX86_ROOT}/src/wrapped/wrappeddbusmenugtk.c"
     "${BOX86_ROOT}/src/wrapped/wrappeddbusmenuglib.c"
     "${BOX86_ROOT}/src/wrapped/wrappedappindicator.c"
@@ -418,6 +473,7 @@ set(WRAPPEDS
     "${BOX86_ROOT}/src/wrapped/wrappedpangocairo.c"
     "${BOX86_ROOT}/src/wrapped/wrappedgthread2.c"
     "${BOX86_ROOT}/src/wrapped/wrappedcairo.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedcairogobject.c"
     "${BOX86_ROOT}/src/wrapped/wrappedpango.c"
     "${BOX86_ROOT}/src/wrapped/wrappeddbusglib1.c"
     "${BOX86_ROOT}/src/wrapped/wrappedgudev1.c"
@@ -462,6 +518,7 @@ set(WRAPPEDS
     "${BOX86_ROOT}/src/wrapped/wrappedpcap.c"
     "${BOX86_ROOT}/src/wrapped/wrappedfaudio.c"
     "${BOX86_ROOT}/src/wrapped/wrappedtbbmallocproxy.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibfuse.c"
 )
 if(ANDROID)
     list(APPEND WRAPPEDS
@@ -515,7 +572,9 @@ if(ARM_DYNAREC)
 
         "${BOX86_ROOT}/src/dynarec/dynarec_arm.c"
         "${BOX86_ROOT}/src/dynarec/dynarec_arm_functions.c"
+        "${BOX86_ROOT}/src/dynarec/dynarec_arm_jmpnext.c"
         "${BOX86_ROOT}/src/dynarec/arm_printer.c"
+        "${BOX86_ROOT}/src/emu/x86test.c"
 
         "${BOX86_ROOT}/src/dynarec/arm_prolog.S"
         "${BOX86_ROOT}/src/dynarec/arm_epilog.S"
@@ -563,6 +622,8 @@ if(ARM_DYNAREC)
     set_target_properties(arm_pass2 PROPERTIES COMPILE_FLAGS "-DSTEP=2")
     add_library(arm_pass3 OBJECT ${DYNAREC_PASS})
     set_target_properties(arm_pass3 PROPERTIES COMPILE_FLAGS "-DSTEP=3")
+    add_library(test_interpreter OBJECT ${INTERPRETER})
+    set_target_properties(test_interpreter PROPERTIES COMPILE_FLAGS "-DTEST_INTERPRETER")
     add_dependencies(arm_pass0 WRAPPERS)
     add_dependencies(arm_pass1 WRAPPERS)
     add_dependencies(arm_pass2 WRAPPERS)
@@ -574,6 +635,7 @@ if(ARM_DYNAREC)
         $<TARGET_OBJECTS:arm_pass1>
         $<TARGET_OBJECTS:arm_pass2>
         $<TARGET_OBJECTS:arm_pass3>
+        $<TARGET_OBJECTS:test_interpreter>
     )
 
 endif()
@@ -583,16 +645,18 @@ if(ARM_DYNAREC)
     add_custom_command(
         OUTPUT "${BOX86_ROOT}/src/git_head.h"
         COMMAND bash -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\">\"${BOX86_ROOT}/src/git_head.h\""
-        DEPENDS dynarec ${ELFLOADER_SRC} ${WRAPPEDS}
+        DEPENDS dynarec ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS}
         VERBATIM)
 else()
     add_custom_command(
         OUTPUT "${BOX86_ROOT}/src/git_head.h"
         COMMAND bash -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\">\"${BOX86_ROOT}/src/git_head.h\""
-        DEPENDS ${ELFLOADER_SRC} ${WRAPPEDS}
+        DEPENDS ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS}
         VERBATIM)
 endif()
 
+add_library(interpreter OBJECT ${INTERPRETER})
+
 if(BOX86LIB)
 #build a static or dynamic lib...
 if(BOX86DYNAMIC)
@@ -602,9 +666,9 @@ else()
 endif()
 if(ARM_DYNAREC)
     if(ANDROID)
-        target_link_libraries(${BOX86} dynarec m dl)
+        target_link_libraries(${BOX86} dynarec interpreter m dl)
     else()
-        target_link_libraries(${BOX86} dynarec m dl rt pthread resolv)
+        target_link_libraries(${BOX86} dynarec interpreter m dl rt pthread resolv)
     endif()
 endif()
 if(${CMAKE_VERSION} VERSION_LESS "3.13")
@@ -630,7 +694,9 @@ else()
     set_target_properties(${BOX86} PROPERTIES LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -Wl,--as-needed -pthread")
 endif()
 if(ARM_DYNAREC)
-    target_link_libraries(${BOX86} dynarec)
+    target_link_libraries(${BOX86} dynarec interpreter)
+else()
+    target_link_libraries(${BOX86} interpreter)
 endif()
 
 if(${CMAKE_VERSION} VERSION_LESS "3.13")
@@ -649,12 +715,17 @@ endif()
 if(NOT _x86 AND NOT _x86_64)
   install(TARGETS ${BOX86}
     RUNTIME DESTINATION bin)
-  configure_file(system/box86.conf.cmake system/box86.conf)
-  install(FILES ${CMAKE_BINARY_DIR}/system/box86.conf DESTINATION /etc/binfmt.d/)
-  install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.6 DESTINATION /usr/lib/i386-linux-gnu/)
-  install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.5 DESTINATION /usr/lib/i386-linux-gnu/)
-  install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libgcc_s.so.1 DESTINATION /usr/lib/i386-linux-gnu/)
-  install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libpng12.so.0 DESTINATION /usr/lib/i386-linux-gnu/)
+  if(NOT NO_CONF_INSTALL)
+    configure_file(system/box86.conf.cmake system/box86.conf)
+    install(FILES ${CMAKE_BINARY_DIR}/system/box86.conf DESTINATION /etc/binfmt.d/)
+    install(FILES ${CMAKE_SOURCE_DIR}/system/box86.box86rc DESTINATION /etc/)
+  endif()
+  if(NOT NO_LIB_INSTALL)
+    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.6 DESTINATION /usr/lib/i386-linux-gnu/)
+    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.5 DESTINATION /usr/lib/i386-linux-gnu/)
+    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libgcc_s.so.1 DESTINATION /usr/lib/i386-linux-gnu/)
+    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libpng12.so.0 DESTINATION /usr/lib/i386-linux-gnu/)
+  endif()
 endif()
 
 if(NOT TARGET uninstall)
@@ -784,7 +855,7 @@ add_test(NAME sse COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/$
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref17.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-set_tests_properties(sse PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTNAN=0")
+set_tests_properties(sse PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTNAN=0;BOX86_DYNAREC_FASTROUND=0")
 
 add_test(NAME longjumpInSignals COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
     -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test18 -D TEST_OUTPUT=tmpfile18.txt 
@@ -810,4 +881,11 @@ foreach(file ${extension_tests})
         -P ${CMAKE_SOURCE_DIR}/runTest.cmake)
 endforeach()
 
+add_test(NAME sse_optimized COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test17_o2 -D TEST_OUTPUT=tmpfile17_o2.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref17_o2.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+set_tests_properties(sse_optimized PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTNAN=0;BOX86_DYNAREC_FASTROUND=0")
+
 endif(BOX86LIB)

debian/changelog

@@ -1,3 +1,64 @@
+box86 (0.3.0) unstable; urgency=low
+
+ * Added rcfile handling: box86 now read "/etc/box86.box86rc" and "~/.box86rc" and apply per process parameters
+ * Added BOX86_DYNAREC_FORWARD to allow bigger Dynarec block to be build
+ * Refactored Strong Memory Model emulation (faster and more accurate)
+ * Box86 now return a custom name for the CPU instal of P4, build with Box86 and the cpu name it's running on
+ * Improved memory protection tracking and signal handling, again
+ * More wrapped libs (especially on the gtk familly)
+ * More wrapped functions
+ * Added support for CreateDeviceEx in d3dadapter9, for better gallium nine support
+ * Improved dladdr/dladdr1 wrapped function
+ * Some fixes to flags handling in the Dynarec on some special cases
+ * Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+ * Lots of improvment in the Rounding of float & double (Dynarec and Interpretor), and proper handling of denormals when converting to/from 80 bits long double
+ * Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+
+ -- Sebastien Chevalier <[email protected]>  Thu, 10 Mar 2023 10:20:33 -0500
+
+box86 (0.2.8) unstable; urgency=low
+
+  * Wrapped more Vulkan functions, dxvk2.0 now works
+  * Added support for overridden malloc/free functions
+  * Improved elf loader
+  * Improved dlsym handling with RTLD_NEXT
+  * Added BOX86_DYNAREC_SAFEFLAGS to allow finetuning Dynarec generated code handling of flags accross function calls and returns.
+  * Added BOX86_BASH to setup x86 bash binary, so shell script can be run in x86 world
+  * Added BOX86_ROLLING_LOG to have details log only on crash
+  * Improved TLS Size handling (Unity3D games now works)
+  * Improved execv family of function to have better box86 integration
+  * Added a few more wrapped libs, like FAudio
+  * Improved gtk, SDL2, GL, vulkan (and more) wrapping
+  * Improved Signal Handling and memory protection traking
+  * Added some basic support for statically linked program. Thread are not working yet, but memory allocation and file access does.
+  * add LX2160A config by @lizthegrey in #612
+  * Grammar and formatting COMPILE.md by @Radiicall in #615
+  * Update README.md by @everypizza1 in #617
+  * docs: add instructions for Debian-based Linux by @ryanfortner in #622
+  * Fix typoes in readme by @ryanfortner in #621
+  * Various improvements to install_steam.sh by @ryanfortner in #620
+  * docs: remove unnecessary newline by @ryanfortner in #624
+  * fix: the first parameter of realpath may be null by @ye-yeshun in #630
+  * Add 6 wrapped function to libdrm, and 2 to libexpat. by @cjy880927 in #632
+  * Fix #610 by @rajdakin in #634
+  * some floating point instructions by @ye-yeshun in #637
+  * Fix Typo in readme x86_64 -> x86 by @Jai-JAP in #642
+  * Fix grammar by @Jai-JAP in #645
+  * Test wine preload by @ptitSeb in #649
+  * Align address when call mprotect by @wannacu in #651
+  * Added 66 0F 7D HSUBPD opcode by @wannacu in #652
+  * add TEGRAX1 and TEGRAX1ARM32 targets on ubuntu-18.04 by @theofficialgman in #654
+  * automatically install 32 bit libraries if 64 bit OS is detected in install_steam.sh by @Moldytzu in #659
+  * mention gcc-arm-linux-gnueabihf in compile.md by @Moldytzu in #660
+  * Add libc6:armhf install to RPi4 aarch64 build info by @WheezyE in #685
+  * Updated wine install for RPiOS bullseye 32/64bit by @WheezyE in #689
+  * Add atomic add opcodes by @Trass3r in #695
+  * steam: disable modern login and pass arguments to client by @Moldytzu in #700
+  * Modify sizeof(instsize_t) to 1 byte by @liuli0217 in #721
+  * Add MOV GS:od, AL by @liuli0217 in #722
+
+ -- Sebastien Chevalier <[email protected]>  Thu, 17 Nov 2022 11:20:33 -0500
+
 box86 (0.2.6) unstable; urgency=low
 
   * Added an option to get generated -NAN on SSE2 operation

debian/rules

@@ -6,7 +6,8 @@
 override_dh_auto_configure:
 	dh_auto_configure -- \
 	      -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-	      -DARM_DYNAREC=1
+	      -DARM_DYNAREC=1 \
+		  -DNOGIT=1
 
 override_dh_shlibdeps:
 	dh_shlibdeps --exclude=libgcc_s.so.1 --exclude=libpng12.so.0 --exclude=libstdc++.so.5 --exclude=libstdc++.so.6 

docs/CHANGELOG.md

@@ -1,3 +1,21 @@
+v0.3.0
+======
+* Added rcfile handling: box86 now read `/etc/box86.box86rc` and `~/.box86rc` and apply per process parameters
+    => Every program can be finetuned, for speed or just to make them work
+* Added BOX86_DYNAREC_FORWARD to allow bigger Dynarec block to be build
+* Refactored Strong Memory Model emulation (faster and more accurate)
+* Box86 now return a custom name for the CPU instal of P4, build with Box86 and the cpu name it's running on
+* Improved memory protection tracking and signal handling, again
+* More wrapped libs (especially on the gtk familly)
+* More wrapped functions
+* Added support for CreateDeviceEx in d3dadapter9, for better gallium nine support
+* Improved dladdr/dladdr1 wrapped function
+* Some fixes to flags handling in the Dynarec on some special cases
+* Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+* Lots of improvment in the Rounding of float & double (Dynarec and Interpretor), and proper handling of denormals when converting to/from 80 bits long double
+* Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+
+
 v0.2.8
 ======
 * Wrapped more vulkan function, dxvk 2.0 now works

docs/COMPILE.md

@@ -6,10 +6,12 @@ You can use [@Itai-Nelken](https://github.com/Itai-Nelken)'s apt repository to i
 
 ```
 sudo wget https://itai-nelken.github.io/weekly-box86-debs/debian/box86.list -O /etc/apt/sources.list.d/box86.list
-wget -qO- https://itai-nelken.github.io/weekly-box86-debs/debian/KEY.gpg | sudo apt-key add -
+wget -qO- https://itai-nelken.github.io/weekly-box86-debs/debian/KEY.gpg | sudo gpg --dearmor -o /etc/apt/trusted.gpg.d/box86-debs-archive-keyring.gpg
 sudo apt update && sudo apt install box86 -y
 ```
 
+**Note:** On a 64bit OS, install the `box86:armhf` package.
+
 Alternatively, you can generate your own package using the [instructions below](https://github.com/ptitSeb/box86/blob/master/docs/COMPILE.md#debian-packaging). 
 
 #### for Pandora
@@ -52,7 +54,7 @@ _Build box86:armhf on RPiOS 64-bit ARM (aarch64)_
 sudo apt install gcc-arm-linux-gnueabihf # building 32-bit ARM code on aarch64 requires this armhf gcc cross-compiler toolchain 
 git clone https://github.com/ptitSeb/box86
 cd box86
-mkdir build; cd build; cmake .. -DRPI4ARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo # -DRPI4ARM64=1 for Pi4 aarch64 (use `-DRPI2ARM64=1` etc for other models)
+mkdir build; cd build; cmake .. -DRPI4ARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo # -DRPI4ARM64=1 for Pi4 aarch64 (use `-DRPI3ARM64=1` for a PI3 model)
 make -j2
 sudo make install
 sudo systemctl restart systemd-binfmt
@@ -70,6 +72,22 @@ As most RK3399 devices run an AARCH64 OS, you'll need an `armhf` multiarch envir
 
 Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
 
+#### for RK3588 / RK3588S
+
+`mkdir build; cd build; cmake .. -DRK3588=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`
+
+As most RK3588 devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for ODROID N2/N2+
+
+`mkdir build; cd build; cmake .. -DODROIDN2=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+As most ODROID N2/N2+ devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
 #### for Tinker Board (1/1S) or RK3288
 
 `mkdir build; cd build; cmake .. -DRK3288=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
@@ -94,12 +112,25 @@ As most Snapdragon 845 devices run an AARCH64 OS, you'll need an `armhf` multiar
 
 As most Phytium (D2000 or FT2000/4) devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for Other ARM64 64bits Linux platform
+
+`mkdir build; cd build; cmake .. -DARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+You can also add `-DBAD_SIGNAL=ON` to the cmake command if you are on Linux Kernel mixed with Android, like on RK3588 or maybe Termux
+
+You'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
 Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
 
 #### for Other ARM Linux platforms
 
  `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
 
+You can also add `-DBAD_SIGNAL=ON` to the cmake command if you are on Linux Kernel mixed with Android, like on RK3588 or maybe Termux
+
 #### for x86 Linux
 
  `mkdir build; cd build; cmake .. -DCMAKE_C_FLAGS="-m32" -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
@@ -163,5 +194,5 @@ NVIDIA doesn't provide armhf libraries for their GPU drivers at this time. There
 
 Debian Packaging
 ----
-Box86 can also be packaged into a .deb file with `DEB_BUILD_OPTIONS=nostrip dpkg-buildpackage -us -uc -nc`.
+Box86 can also be packaged into a .deb file ***using the source code zip from the releases page*** with `DEB_BUILD_OPTIONS=nostrip dpkg-buildpackage -us -uc -nc`. Configure any additional cmake options you might want in `debian/rules`.
 

docs/README.md

@@ -12,9 +12,9 @@ Box86 lets you run x86 Linux programs (such as games) on non-x86 Linux systems,
 
 You *NEED* a 32-bit subsystem to run and build Box86. Box86 is useless on 64-bit only systems. Also, you *NEED* a 32-bit toolchain to build Box86. A toolchain that only supports 64-bit will not compile Box86, and you'll get errors (typically on aarch64, you get "-marm" not recognized, and you'll need a multiarch or chroot environment).
 
-Because Box86 uses the native versions of some "system" libraries, like libc, libm, SDL, and OpenGL, it's easy to integrate and use with most applications, and performance can be surprisingly high in many cases. Take a look at those bench analysis for an example [here](https://box86.org/index.php/2021/06/game-performances/). That also means that you will need a 32bits userspace on 64bits OS, like `armhf` on top an ̀`aarch64` 64bits OS.
+Because Box86 uses the native versions of some "system" libraries, like libc, libm, SDL, and OpenGL, it's easy to integrate and use with most applications, and performance can be surprisingly high in many cases. Take a look at those bench analysis for an example [here](https://box86.org/index.php/2021/06/game-performances/). That also means that you will need a 32bits userspace on 64bits OS, like `armhf` on top an `aarch64` 64bits OS.
 
-Most x86 Games need OpenGL, so on ARM platforms a solution like [gl4es](https://github.com/ptitSeb/gl4es) might be necessary. (Some ARM platforms only support OpenGL ES and/or their OpenGL implementation is dodgy (see OpenGL on Android).)
+Most x86 Games need OpenGL, so on ARM platforms a solution like [gl4es](https://github.com/ptitSeb/gl4es) might be necessary. (Some ARM platforms only support OpenGL ES and/or their OpenGL implementation is dodgy. (see OpenGL on Android))
 
 Box86 now integrates a DynaRec (dynamic recompiler) for the ARM platform, providing a speed boost between 5 to 10 times faster than only using the interpreter. Some high level information on how the Dynarec work can be found [here](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/).
 
@@ -76,6 +76,15 @@ If you look at a 64bits version of box86, look at [Box64](https://github.com/pti
 
 ----
 
+Notes about Box86 configuration
+----
+
+Box86 now have configurations files. There are 2 files loaded. `/etc/box4.box86rc` and `~/.box86rc`. Both files have the same syntax, and is basicaly an ini files. Section in square brakets define the process name, and the rest is the env. var. to set. Looke at [Usage](USAGE.md) for detail on what parameters can be put. Box86 comes with a default file that should be installed for better stability. The file in in `system/box86.box86rc` and should be installed to `/etc/box86.box86rc` If, for some reasons, you don't want to install that file here, at least copy it to `~/.box86rc` or some game may not function correctly.
+Note that the priority is: `~/.bashrc` > `/etc/box86.box86rc` > command line
+So, your settings in `~/.bashrc` may override the setting from your command line...
+
+----
+
 Notes about Unity game emulation
 ----
 
@@ -87,15 +96,20 @@ Hint: on Pi4, use `MESA_GL_VERSION_OVERRIDE=3.2` and with Panfrost use `PAN_MESA
 Notes about Steam
 ----
 
-Linux Steam's can run now with box86. But it's still a bit unstable, and not everything works:
-- First problem is Steam crashing after the sign-in window, if you encounter this issue, you may need to add libappindicator. To install it on Debian, run `sudo apt install libappindicator1`.
-- If you select to "Remember password", Steam is crashing on subsequent starts, unless you have libnm installed. To install it on Debian, run `sudo apt install libnm0`.
+Linux Steam's can run now with box86. but you also need box64 for it to completly usable.
+It is advised to run steam in the Small mode, as it use the less memory, but steamwebhelper (a 64bits process) will still be loaded even if not used.
+The login screen cannot run without steamwebhelper, and will just show a blank windows without box64 properly setup in the system.
+File note, Steam will use a lot of memory, and barely fit on system with 4GB of RAM. It will not work anymore on system with less memory (as a workaround, create a swap file, login and check "remember me", the use box64rc to disable steamwebhelper and run only in small mode without swap after the 1st login)
+Final note: the Steam BigPicture will work, but also need steamwebhelper (and so box64), and lots of memory. It will not start on system with only 4GB of RAM without swap.
 - If you have trouble installing Steam, you can find `install_steam.sh` in the root folder of the box86 repo. This simple script will download and install steam in your home folder, and then create a shortcut to steam in `/usr/local/bin` (and for this it will ask for sudo permission). Simply use `steam` to launch once it's installed. Note that the installation, being in the Home folder, will only work for a single user. Don't use this script if you need a multi-user installation.
-- Once open, Steam will only work on "Small Mode" and in "Big Picture", not in the regular "Large Mode". This is because some Steam components used in the browser view are only 64-bit now. So go in the "View" menu and switch to "Small view", else the list will stay empty. Alternatively, Steam can be started in small mode directly by using `+open steam://open/minigameslist` command line arguments.
 - To avoid the "libc.so.6 is absent" message, you can use `STEAMOS=1` and `STEAM_RUNTIME=1` as environment variables (it's automatically there if you used the `install_steam.sh` script)
-- Some Steam games (most Source engine games, like "Portal" or "Half-Life 2") use libtcmalloc. Box86 will detect it and will try to LD_PRELOAD it, for better compatibility. While it should work without the aforementioned feature, it is safer to add it to your system if you intend to play those games. To install it on Debian, run `sudo apt install libtcmalloc-minimal4`.
 
-Steam for Windows installs fine but doesn't work yet.
+to disable `steamwebhelper` when box64 installed and running, create (or edit) `~/.box64rc` and add:
+```
+[steamwebhelper]
+BOX64_EXIT=1
+```
+in it and it will be disable. Comment with a `#` those 2 lines to enable it again
 
 ----
 

docs/USAGE.md

@@ -3,20 +3,21 @@ Usage
 ----
 
 There are many environment variables to control Box86 behaviour. 
+Env. var with * can also be put inside box64rc files
 
-#### BOX86_LOG
+#### BOX86_LOG *
 Controls the Verbosity level of the logs
  * 0: NONE : No message (except some fatal error). (Default.)
  * 1: INFO : Show some minimum log (Example: librairies not found)
  * 2: DEBUG : Details a lot of stuff (Example: relocations or functions called).
  * 3: DUMP : All DEBUG plus DUMP of all ELF Info (so same as `BOX86_LOG=2 BOX86_DUMP=1`)
 
-#### BOX86_DUMP
+#### BOX86_DUMP *
 Controls the Dump of elf content
  * 0: No dump of elf information (Default)
  * 1: Dump elf sections and relocations and other information (also active if using `BOX86_LOG=3`)
 
-#### BOX86_ROLLING_LOG
+#### BOX86_ROLLING_LOG *
 Show last few wrapped function call when a Signal is caught
  * 0: No last function call printed (Default.)
  * 1: Last 16 wrapped functions calls printed when a signal is printed. Incompatible with BOX86_LOG>1 (may need BOX86_SHOWSEGV=1 also)
@@ -27,173 +28,223 @@ Disables Box86 printing its version and build
  * 0 : Enable printing its banner. (Default.)
  * 1 : Disable printing its banner. 
 
-#### BOX86_LD_LIBRARY_PATH
+#### BOX86_LD_LIBRARY_PATH *
 Path to look for x86 libraries. Default is current folder and `lib` in current folder.
 Also, `/usr/lib/i386-linux-gnu` and `/lib/i386-linux-gnu` are added if they exist.
 
-#### BOX86_PATH
+#### BOX86_PATH *
 Path to look for x86 executable. Default is current folder and `bin` in current folder.
 
-#### BOX86_DLSYM_ERROR
+#### BOX86_DLSYM_ERROR *
 Enables/Disables the logging of `dlsym` errors.
  * 0 : Don't log `dlsym` errors. (Default.)
  * 1 : Log dlsym errors.
 
-#### BOX86_TRACE_FILE
+#### BOX86_TRACE_FILE *
 Send all log and trace to a file instead of `stdout`
 Also, if name contains `%pid` then this is replaced by the actual PID of box86 instance
 Use `stderr` to use this instead of default `stdout` 
 
-#### BOX86_TRACE
+#### BOX86_TRACE *
 Only on build with trace enabled. Trace allow the logging of all instruction executed, along with register dump
  * 0 : No trace. (Default.) 
  * 1 : Trace enabled. Trace start after the initialisation of all depending libraries is done.
  * symbolname : Trace only `symbolname` (trace is disable if the symbol is not found).
  * 0xXXXXXXX-0xYYYYYYY : Trace only between the 2 addresses.
 
-#### BOX86_TRACE_INIT
+#### BOX86_TRACE_INIT *
 Use BOX86_TRACE_INIT instead of BOX_TRACE to start trace before the initialisation of libraries and the running program
  * 0 : No trace. (Default.)
  * 1 : Trace enabled. The trace start with the initialisation of all depending libraries is done.
 
-#### BOX86_TRACE_START
+#### BOX86_TRACE_START *
 Only on builds with trace enabled.
  * NNNNNNN : Start trace only after NNNNNNNN opcode execute (number is an `uint64_t`).
 
-#### BOX86_TRACE_XMM
+#### BOX86_TRACE_XMM *
 Only on builds with trace enabled.
  * 0 : The XMM (i.e. SSE/SSE2) register will not be logged with the general and x86 registers. (Default.)
  * 1 : Dump the XMM registers.
 
-#### BOX86_LOAD_ADDR
+#### BOX86_TRACE_EMM *
+Only on builds with trace enabled.
+ * 0 : The EMM (i.e. MMX) register will not be logged with the general and x86 registers. (Default.)
+ * 1 : Dump the EMM registers.
+
+#### BOX86_LOAD_ADDR *
 Try to load at 0xXXXXXX main binary (if binary is a PIE)
  * 0xXXXXXXXX : The load address . (Only active on PIE programs.)
 
-#### BOX86_NOSIGSEGV
+#### BOX86_NOSIGSEGV *
 Disable handling of SigSEGV. (Very useful for debugging.)
  * 0 : Let the x86 program set sighandler for SEGV (Default.)
  * 1 : Disable the handling of SigSEGV.
 
-#### BOX86_NOSIGILL
+#### BOX86_NOSIGILL *
 Disable handling of SigILL (to ease debugging mainly).
  * 0 : Let x86 program set sighandler for Illegal Instruction
  * 1 : Disables the handling of SigILL 
 
-#### BOX86_X11COLOR16
+#### BOX86_X11COLOR16 *
 PANDORA only: Try converting X11 color from 32 bits to 16 bits (to avoid light green on light cyan windows).
  * 0 : Don't touch X11 colors. (Default.)
  * 1 : Change colors arguments in XSetForeground, XSetBackground and XCreateGC.
 
-#### BOX86_X11THREADS
+#### BOX86_X11THREADS *
 Call XInitThreads when loading X11. (This is mostly for old Loki games with the Loki_Compat library.)
  * 0 : Don't force call XInitThreads. (Default.)
  * 1 : Call XInitThreads as soon as libX11 is loaded.
 
-#### BOX86_X11GLX
+#### BOX86_X11GLX *
 Force libX11's GLX extension to be present.
 * 0 : Do not force libX11's GLX extension to be present. 
 * 1 : GLX will always be present when using XQueryExtension. (Default.)
 
-#### BOX86_DYNAREC_DUMP
+#### BOX86_DYNAREC_DUMP *
 Enables/disables Box86's Dynarec's dump.
  * 0 : Disable Dynarec's blocks dump. (Default.)
  * 1 : Enable Dynarec's blocks dump.
  * 2 : Enable Dynarec's blocks dump with some colors.
 
-#### BOX86_DYNAREC_LOG
+#### BOX86_DYNAREC_LOG *
 Set the level of DynaRec's logs.
  * 0 : NONE : No Logs for DynaRec. (Default.)
  * 1 :INFO : Minimum Dynarec Logs (only unimplemented OpCode).
  * 2 : DEBUG : Debug Logs for Dynarec (with details on block created / executed).
  * 3 : VERBOSE : All of the above plus more.
 
-#### BOX86_DYNAREC
+#### BOX86_DYNAREC *
 Enables/Disables Box86's Dynarec.
  * 0 : Disables Dynarec.
  * 1 : Enable Dynarec. (Default.)
 
-#### BOX86_DYNAREC_TRACE
+#### BOX86_DYNAREC_TRACE *
 Enables/Disables trace for generated code.
  * 0 : Disable trace for generated code. (Default.)
  * 1 : Enable trace for generated code (like regular Trace, this will slow down the program a lot and generate huge logs).
 
-#### BOX86_NODYNAREC 
+#### BOX86_NODYNAREC *
 Forbid dynablock creation in the interval specified (helpfull for debugging behaviour difference between Dynarec and Interpretor)
  * 0xXXXXXXXX-0xYYYYYYYY : define the interval where dynablock cannot start (inclusive-exclusive)
 
-#### BOX86_DYNAREC_BIGBLOCK
+#### BOX86_DYNAREC_TEST *
+Dynarec will compare it's execution with the interpreter (super slow, only for testing)
+ * 0 : No comparison
+ * 1 : Each opcode runs on interepter and on Dynarec, and regs and memory are compared and print if different
+
+#### BOX86_DYNAREC_BIGBLOCK *
 Enables/Disables Box86's Dynarec building BigBlock.
  * 0 : Don't try to build block as big as possible (can help program using lots of thread and a JIT, like C#/Unity) (Default when libmonobdwgc-2.0.so is loaded)
  * 1 : Build Dynarec block as big as possible (Default.)
  * 2 : Build Dynarec block bigger (don't stop when block overlaps)
 
-#### BOX86_DYNAREC_STRONGMEM
+#### BOX86_DYNAREC_FORWARD *
+Define Box86's Dynarec max allowed forward value when building Block.
+ * 0 : No forward value. When current block end, don't try to go further even if there are previous forward jumps
+ * XXX : Allow up to XXXX bytes of gap when building a Block after the block end to next forward jump (Default: 128)
+ 
+#### BOX86_DYNAREC_STRONGMEM *
 Enable/Disable simulation of Strong Memory model
 * 0 : Don't try anything special (Default.)
 * 1 : Enable some Memory Barrier when reading from memory (on some MOV opcode) to simulate Strong Memory Model while trying to limit performance impact (Default when libmonobdwgc-2.0.so is loaded)
 * 2 : Enable some Memory Barrier when reading from memory (on some MOV opcode) to simulate Strong Memory Model
 
-#### BOX86_DYNAREC_X87DOUBLE
+#### BOX86_DYNAREC_X87DOUBLE *
 Force the use of Double for x87 emulation
 * 0 : Try to use float when possible for x87 emulation (faster, Default)
 * 1 : Only use Double for x87 emulation (slower, may be needed for some specific games, like Crysis were it's enable automatically)
 
-#### BOX86_DYNAREC_FASTNAN
+#### BOX86_DYNAREC_FASTNAN *
 Enable/Disable generation of -NAN
 * 0 : Generate -NAN like on x86 (slower, more x86 accurate)
 * 1 : Don't do anything special with NAN, to go as fast as possible (Default)
 
-#### BOX86_DYNAREC_SAFEFLAGS
+#### BOX86_DYNAREC_FASTROUND *
+Enable/Disable generation of precise x86 rounding
+* 0 : Generate float/double -> int rounding like on x86
+* 1 : Don't do anything special with edge case Rounding, to go as fast as possible (no INF/NAN/Overflow -> MIN_INT conversion) (faster, Default)
+
+#### BOX86_DYNAREC_SAFEFLAGS *
 Handling of flags on CALL/RET opcodes
 * 0 : Treat CALL/RET as if it never needs any flags (faster but may have side-effects)
 * 1 : most of RET will need flags, most of CALLS will not (Default)
 * 2 : All CALL/RET will need flags (slower, but might be needed. Automatically enabled for Vara.exe)
 
-#### BOX86_LIBGL
+#### BOX86_DYNAREC_HOTPAGE *
+Handling of HotPage (Page beeing both executed and writen)
+* 0 : Don't track hotpage
+* 1-255 : Trak HotPage, and disable execution of a page beeing writen for N attempts (default is 16)
+
+#### BOX86_DYNAREC_FASTPAGE *
+Will use a faster handling of HotPage (Page being both executed and written)
+* 0 : use regular hotpage (Default)
+* 1 : Use faster hotpage, taking the risk of running obsolete JIT code (might be faster, but more prone to crash)
+
+#### BOX86_DYNAREC_BLEEDING_EDGE *
+Detect MonoBleedingEdge and apply conservative settings
+* 0 : Don't detect MonoBleedingEdge
+* 1 : Detect MonoBleedingEdge, and apply BIGBLOCK=0 STRONGMEM=1 if detected (Default)
+
+#### BOX86_DYNAREC_WAIT *
+Behavior with FillBlock is not availble (FillBlock build Dynarec blocks and is not multithreaded)
+* 0 : Dynarec will not wait for FillBlock to ready and use Interpreter instead (might speedup a bit massive multithread or JIT programs)
+* 1 : Dynarec will wait for FillBlock to be ready (Default)
+
+#### BOX86_SSE_FLUSHTO0 *
+Handling of SSE Flush to 0 flags
+* 0 : Just track the flag (Default)
+* 1 : Direct apply of SSE Flush to 0 flag
+
+#### BOX86_X87_NO80BITS *
+Handling of x87 80bits long double
+* 0 : Try to handle 80bits long double as precise as possible (Default)
+* 1 : Handle them as double
+
+#### BOX86_LIBGL *
  * libXXXX set the name for libGL (defaults to libGL.so.1).
  * /PATH/TO/libGLXXX : Sets the name and path for libGL
  You can also use SDL_VIDEO_GL_DRIVER
 
-#### BOX86_LD_PRELOAD
+#### BOX86_LD_PRELOAD *
  * XXXX[:YYYYY] force loading XXXX (and YYYY...) libraries with the binary
- PreLoaded libs can be emulated or native, and are treated the same way as if they were comming from the binary
+ PreLoaded libs can be emulated or native, and are treated the same way as if they were comming from the binary (not that using this inside a rcfile might not work if the process needs to be renamed, for example for Wine process)
  
-#### BOX86_EMULATED_LIBS
+#### BOX86_EMULATED_LIBS *
  * XXXX[:YYYYY] force lib XXXX (and YYYY...) to be emulated (and not wrapped)
 Some games uses an old version of some libraries, with an ABI incompatible with native version.
 Note that LittleInferno for example is auto detected, and libvorbis.so.0 is automatical added to emulated libs, and same for Don't Starve (and Together / Server variant) that use an old SDL2 too
 
-#### BOX86_ALLOWMISSING_LIBS
+#### BOX86_ALLOWMISSING_LIBS *
 Allow Box86 to continue even if a library is missing (renamed from BOX86_ALLOWMISSINGLIBS). 
  * 0 : Box86 will stop if a library cannot be loaded. (Default.)
  * 1 : Continue even if a needed library cannot be loaded. Unadvised, this will, in most cases, crash later on.
 
-#### BOX86_ALLOWMISSING_SYMBOLS
+#### BOX86_ALLOWMISSING_SYMBOLS *
 Allow dlopen with RTLD_NOW flag to not resolve all symbols. 
  * 0 : Box86 will stop if a library have unresolved symbols. (Default.)
  * 1 : Continue even if a needed library have unresolved symbols (like symbols not yet wrapped).
 
-#### BOX86_PREFER_WRAPPED
+#### BOX86_PREFER_WRAPPED *
 Box86 will use wrapped libs even if the lib is specified with absolute path
  * 0 : Try to use emulated libs when they are defined with absolute path  (Default.)
  * 1 : Use Wrapped native libs even if path is absolute
 
-#### BOX86_PREFER_EMULATED
+#### BOX86_PREFER_EMULATED *
 Box86 will prefer emulated libs first (execpt for glibc, alsa, pulse, GL, vulkan and X11
  * 0 : Native libs are prefered (Default.)
  * 1 : Emulated libs are prefered (Default for program running inside pressure-vessel)
 
-#### BOX86_NOPULSE
+#### BOX86_NOPULSE *
 Disables the load of pulseaudio libraries.
  * 0 : Load pulseaudio libraries if found. (Default.)
  * 1 : Disables the load of pulse audio libraries (libpulse and libpulse-simple), both the native library and the x86 library
 
-#### BOX86_NOGTK
+#### BOX86_NOGTK *
 Disables the loading of wrapped GTK libraries.
  * 0 : Load wrapped GTK libraries if found. (Default.)
  * 1 : Disables loading wrapped GTK libraries.
 
-#### BOX86_NOVULKAN
+#### BOX86_NOVULKAN *
 Disables the load of vulkan libraries.
  * 0 : Load vulkan libraries if found.
  * 1 : Disables the load of vulkan libraries, both the native and the i386 version (can be useful on Pi4, where the vulkan driver is not quite there yet.)
@@ -206,22 +257,22 @@ Disables the load of vulkan libraries.
  * XXX=yyyy
  will add XXX=yyyy env. var. and continue with BOX86_ENV2 ... until var doesn't exist
 
-#### BOX86_FIX_64BIT_INODES
+#### BOX86_FIX_64BIT_INODES *
 Fix/Don't fix 64bit inodes
  * 0 : Don't fix 64bit inodes. (Default.)
  * 1 : Fix 64bit inodes. Helps when running on Filesystems with 64bit inodes. Is useful when a program uses API functions which doesn't support it and the program doesn't use inodes information.
 
-#### BOX86_SHOWSEGV
+#### BOX86_SHOWSEGV *
 Show Segfault signal even if a signal handler is present
  * 0 : Don't force segfault details when a signal handler is install (Default, to limit message spamming)
  * 1 : Show segfault details
 
-#### BOX86_BACKTRACE
+#### BOX86_SHOWBT *
 Show a backtrace (native and emulated) on Signal
  * 0 : Don't show any backtrace (Default.)
  * 1 : Show native backtrace when a signal is dumped (not available on Andoid build)
 
-#### BOX86_BASH
+#### BOX86_BASH *
 Define x86 bash to launch script
  * yyyy
 Will use yyyy as x86 bash to launch script. yyyy needs to be a full path to a valid x86 version of bash. Note that is Box64 is available, an x86_64 version of bash can also be used. Also, BOX64_BASH env. var. will also be read if BOX86_BASH is not defined.
@@ -230,7 +281,8 @@ Will use yyyy as x86 bash to launch script. yyyy needs to be a full path to a va
 Define where to find box64 binary
  * yyyy
 Will use yyyy as x86_64 interpretor, to launch x64_64 binaries
-#### BOX86_JITGDB
+
+#### BOX86_JITGDB *
 
  * 0 : Just print the Segfault message on segfault (default)
  * 1 : Launch `gdb` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process and go in an endless loop, waiting.
@@ -241,3 +293,6 @@ Will use yyyy as x86_64 interpretor, to launch x64_64 binaries
  * 2 : Launch `gdbserver` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process, and go in an endless loop, waiting.
  Use `gdb /PATH/TO/box86` and then `target remote 127.0.0.1:1234` to connect to the gdbserver (or use actual IP if not on the machine). After that, the procedure is the same as with ` BOX86_JITGDB=1`.
  This mode can be usefullwhen programs redirect all console output to a file (like Unity3D Games)
+
+#### BOX86_NORCFILES
+If the env var exist, no rc files (like /etc/box86.box86rc and ~/.box86rc) will be loaded

install_steam.sh

@@ -21,19 +21,28 @@ cd ../ && rm -rf ./tmp/
 echo "#!/bin/bash
 export STEAMOS=1
 export STEAM_RUNTIME=1
+export DBUS_FATAL_WARNINGS=0
 ~/steam/bin/steam -noreactlogin steam://open/minigameslist $@" > steam
 
 # make script executable and move
 chmod +x steam
 sudo mv steam /usr/local/bin/
 
-# detect if we're running on 64 bit Debian 
+# detect if we're running on 64 bit Debian (maybe this can be moved in another script then mentioned in documentation for the people that don't use steam but don't want a headache)
 MACHINE_TYPE=`uname -m`
 if [ ${MACHINE_TYPE} == 'aarch64' ] && [ -f '/etc/debian_version' ]; then
  echo "Detected 64 bit ARM Debian. Installing 32 bit libraries"
  sudo dpkg --add-architecture armhf # enable installation of armhf libraries
  sudo apt update # update package lists with the newly added arch
- sudo apt install libc6:armhf libncurses5:armhf libsdl2*:armhf libopenal*:armhf libpng*:armhf libfontconfig*:armhf libXcomposite*:armhf libbz2-dev:armhf libXtst*:armhf # install the libraries that Steam requires
+ # install the libraries that Steam requires
+ sudo apt install libc6:armhf libncurses5:armhf libsdl2-2.0-0:armhf libsdl2-image-2.0-0:armhf libsdl2-mixer-2.0-0:armhf libsdl2-ttf-2.0-0:armhf libopenal1:armhf libpng16-16:armhf libfontconfig1:armhf libxcomposite1:armhf libbz2-1.0:armhf libxtst6:armhf libsm6:armhf libice6:armhf libgl1:armhf libxinerama1:armhf libxdamage1:armhf
+
+ # install mesa for armhf if already installed
+ if dpkg-query -W libgl1-mesa-dri 2>/dev/null; then
+  echo "Detected mesa driver for 64 bit ARM. Installing complementary 32 bit one"
+  sudo apt install libgl1-mesa-dri:armhf
+ fi
+
  echo "Don't forget to compile/install Box64!"
 fi
 

pkgbuilds/PKGBUILD-a64

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
-pkgname=box86-rk3399-git
-pkgver=r3858.5368285
+pkgname=box86-a64-git
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -9,21 +9,28 @@ url="https://github.com/ptitSeb/box86"
 license=('MIT')
 optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
 makedepends=('git' 'cmake' 'make')
-provides=("${pkgname%-git}" "${pkgname%-rk3399-git}-git" "${pkgname%-rk3399-git}")
-conflicts=("${pkgname%-git}" "${pkgname%-rk3399-git}-git" "${pkgname%-rk3399-git}")
+provides=("${pkgname%-git}" "${pkgname%-a64-git}-git" "${pkgname%-a64-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-a64-git}-git" "${pkgname%-a64-git}")
 source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')
 
 pkgver() {
-	cd "$srcdir/${pkgname%-rk3399-git}"
+	cd "${srcdir}/${pkgname%-a64-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
+prepare() {
+	cd "${srcdir}/${pkgname%-a64-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DA64=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-rk3399-git}" -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-a64-git}/build"
+	make -j$(nproc)
 }
 
 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-a64-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-arm

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-arm-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')
 
 pkgver() {
-	cd "$srcdir/${pkgname%-arm-git}"
+	cd "${srcdir}/${pkgname%-arm-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
+prepare() {
+	cd "${srcdir}/${pkgname%-arm-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-arm-git}" -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-arm-git}/build"
+	make -j$(nproc)
 }
 
 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-arm-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-arm64

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

pkgbuilds/PKGBUILD-gameshell

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-gameshell-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')
 
 pkgver() {
-	cd "$srcdir/${pkgname%-gameshell-git}"
+	cd "${srcdir}/${pkgname%-gameshell-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
+prepare() {
+	cd "${srcdir}/${pkgname%-gameshell-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DGAMESHELL=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-gameshell-git}" -DGAMESHELL=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-gameshell-git}/build"
+	make -j$(nproc)
 }
 
 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-gameshell-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-goaclone

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

pkgbuilds/PKGBUILD-lx2610a

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

pkgbuilds/PKGBUILD-odroid

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-odroid-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')
 
 pkgver() {
-	cd "$srcdir/${pkgname%-odroid-git}"
+	cd "${srcdir}/${pkgname%-odroid-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
+prepare() {
+	cd "${srcdir}/${pkgname%-odroid-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DODROID=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-odroid-git}" -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-odroid-git}/build"
+	make -j$(nproc)
 }
 
 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-odroid-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-odroidn2

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

pkgbuilds/PKGBUILD-pandora

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-pandora-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')
 
 pkgver() {
-	cd "$srcdir/${pkgname%-pandora-git}"
+	cd "${srcdir}/${pkgname%-pandora-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
+prepare() {
+	cd "${srcdir}/${pkgname%-pandora-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DPANDORA=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-pandora-git}" -DPANDORA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-pandora-git}/build"
+	make -j$(nproc)
 }
 
 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-pandora-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-phytium

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

pkgbuilds/PKGBUILD-pyra

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-pyra-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')
 
 pkgver() {
-	cd "$srcdir/${pkgname%-pyra-git}"
+	cd "${srcdir}/${pkgname%-pyra-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
+prepare() {
+	cd "${srcdir}/${pkgname%-pyra-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DPYRA=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-pyra-git}" -DPYRA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-pyra-git}/build"
+	make -j$(nproc)
 }
 
 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-pyra-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-raspberrypi

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-rpi-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -16,11 +16,12 @@ md5sums=('SKIP')
 _pimodel=`tr -d '\0' < /sys/firmware/devicetree/base/model`
 
 pkgver() {
-	cd "$srcdir/${pkgname%-rpi-git}"
+	cd "${srcdir}/${pkgname%-rpi-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
-build() {
+prepare() {
+	cd "${srcdir}/${pkgname%-rpi-git}"
 	# Check for Pi version number
 	if [[ ${_pimodel} =~ "Raspberry Pi 4" ]]; then
 		_piversion=4
@@ -31,15 +32,20 @@ build() {
 	else
 		# This RPI is not supported or this device is not a pi.
 		# We should cancel then the build.
-		echo -e "\033[01m\033[01;31m==> ERROR:\033[00;01m Your device is not supported! This script is for RPi model 4/3/2 (1/0 isn't supported yet).\033[00m"
+		echo -e "\033[01m\033[01;31m==> ERROR:\033[00;01m Your device is not supported! This script is for RPi model 4/3/2 (1/0 isn't supported).\033[00m"
 		exit 1
 	fi
 	echo -e "\033[01m\033[01;34m==> INFO:\033[00;01m Your detected Pi verion number is: \033[01;31m${_piversion}\033[01m.\033[00m"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DRPI${_piversion}=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
 
-	cmake -B build -S "${pkgname%-rpi-git}" -DRPI${_piversion}=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+build() {
+	cd "${srcdir}/${pkgname%-rpi-git}/build"
+	make -j$(nproc)
 }
 
 package() {
-	make -C build DESTDIR="${pkgdir}/" install
+	cd "${srcdir}/${pkgname%-rpi-git}/build"
+	make DESTDIR="${pkgdir}/" install
 }

pkgbuilds/PKGBUILD-rk3xxx

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

pkgbuilds/PKGBUILD-sd845

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

pkgbuilds/PKGBUILD-tegra

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

pkgbuilds/PKGBUILD-x86

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-x86-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('i386')
@@ -14,15 +14,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')
 
 pkgver() {
-	cd "$srcdir/${pkgname%-x86-git}"
+	cd "${srcdir}/${pkgname%-x86-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
+prepare() {
+	cd "${srcdir}/${pkgname%-x86-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DLD80BITS=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-x86-git}" -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-x86-git}/build"
+	make -j$(nproc)
 }
 
 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-x86-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/README.md

@@ -12,7 +12,7 @@ This will copy the PKGBUILD for your platform (where you type your platform name
 In the future, I might try to do an unified PKGBUILD for all platforms that will recognize your platform automatically.
 
 ### 2. What platforms are currently supported.
-All specified in [COMPILE.md](../docs/COMPILE.md).
+Most specified in [COMPILE.md](../docs/COMPILE.md). Except PowerPC based.
 
 ### 3. Any advantages with that sollution?
 This might give you some advantages like those:

rebuild_wrappers.py

@@ -1249,7 +1249,7 @@ def generate_files(root: str, files: Iterable[str], ver: str, gbls: SortedGlobal
 		f.write("void {0}(x86emu_t *emu, uintptr_t fcn) {2} {1} fn = ({1})fcn; ".format(N.orig, W, "{"))
 		
 		if N.orig == 'vFv':
-			f.write("(void)fcn; ")
+			f.write("(void)emu; ")
 		
 		args = N.orig[2:]
 		if args == 'Ev': args = 'E'

src/box86context.c

@@ -22,6 +22,8 @@
 #include <sys/mman.h>
 #include "custommem.h"
 #include "dictionnary.h"
+#include "rcfile.h"
+#include "gltools.h"
 #ifdef DYNAREC
 #include "dynablock.h"
 #include "dynarec/arm_lock_helper.h"
@@ -48,6 +50,7 @@ void finiAllHelpers(box86context_t* context)
     static int finied = 0;
     if(finied)
         return;
+    DeleteParams();
     fini_pthread_helper(context);
     fini_signal_helper();
     fini_bridge_helper();
@@ -82,11 +85,20 @@ int unlockMutex()
 {
     int ret = unlockCustommemMutex();
     int i;
+    #ifdef DYNAREC
+    void* tid = (void*)GetTID();
+    #define GO(A, B)                    \
+        i = (arm_lock_storeifref2(&A, NULL, tid)==tid);  \
+        if(i) {                         \
+            ret|=(1<<B);                \
+        }
+    #else
     #define GO(A, B)                    \
         i = checkUnlockMutex(&A);       \
         if(i) {                         \
             ret|=(1<<B);                \
         }
+    #endif
 
     GO(my_context->mutex_once, 5)
     GO(my_context->mutex_once2, 6)
@@ -100,7 +112,6 @@ int unlockMutex()
     GO(my_context->mutex_thread, 10)
     GO(my_context->mutex_bridge, 11)
     #undef GO
-
     return ret;
 }
 
@@ -108,7 +119,7 @@ void relockMutex(int locks)
 {
     #define GO(A, B)                    \
         if(locks&(1<<B))                \
-            pthread_mutex_lock(&A);     \
+            mutex_lock(&A);             \
 
     GO(my_context->mutex_once, 5)
     GO(my_context->mutex_once2, 6)
@@ -127,22 +138,30 @@ void relockMutex(int locks)
 
 static void init_mutexes(box86context_t* context)
 {
+
+#ifdef DYNAREC
+    arm_lock_stored(&context->mutex_dyndump, 0);
+    arm_lock_stored(&context->mutex_once, 0);
+    arm_lock_stored(&context->mutex_once2, 0);
+    arm_lock_stored(&context->mutex_trace, 0);
+    arm_lock_stored(&context->mutex_tls, 0);
+    arm_lock_stored(&context->mutex_thread, 0);
+    arm_lock_stored(&context->mutex_bridge, 0);
+    pthread_mutex_init(&context->mutex_lock, NULL);
+#else
     pthread_mutexattr_t attr;
     pthread_mutexattr_init(&attr);
     pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK);
     pthread_mutex_init(&context->mutex_once, &attr);
     pthread_mutex_init(&context->mutex_once2, &attr);
     pthread_mutex_init(&context->mutex_trace, &attr);
-#ifdef DYNAREC
-    pthread_mutex_init(&context->mutex_dyndump, &attr);
-#else
-    pthread_mutex_init(&context->mutex_lock, &attr);
-#endif
     pthread_mutex_init(&context->mutex_tls, &attr);
     pthread_mutex_init(&context->mutex_thread, &attr);
     pthread_mutex_init(&context->mutex_bridge, &attr);
+    pthread_mutex_init(&context->mutex_lock, &attr);
 
     pthread_mutexattr_destroy(&attr);
+#endif
 }
 
 static void atfork_child_box64context(void)
@@ -205,9 +224,12 @@ box86context_t *NewBox86Context(int argc)
     context->system = NewBridge();
     context->globaldefver = NewDefaultVersion();
     context->weakdefver = NewDefaultVersion();
+    // Cannot use Bridge name as the map is not initialized yet
     // create vsyscall
     context->vsyscall = AddBridge(context->system, iFEv, x86Syscall, 0, NULL);
     addAlternate((void*)0xffffe400, (void*)context->vsyscall);
+    // create exit bridge
+    context->exit_bridge = AddBridge(context->system, NULL, NULL, 0, NULL);
 #ifdef BUILD_LIB
     context->box86lib = RTLD_DEFAULT;   // not ideal
 #else
@@ -233,7 +255,7 @@ box86context_t *NewBox86Context(int argc)
 
     return context;
 }
-
+void freeALProcWrapper(box86context_t* context);
 EXPORTDYN
 void FreeBox86Context(box86context_t** context)
 {
@@ -342,17 +364,17 @@ void FreeBox86Context(box86context_t** context)
 
     finiAllHelpers(ctx);
 
+#ifdef DYNAREC
+    pthread_mutex_destroy(&ctx->mutex_lock);
+#else
     pthread_mutex_destroy(&ctx->mutex_once);
     pthread_mutex_destroy(&ctx->mutex_once2);
     pthread_mutex_destroy(&ctx->mutex_trace);
-#ifdef DYNAREC
-    pthread_mutex_destroy(&ctx->mutex_dyndump);
-#else
-    pthread_mutex_destroy(&ctx->mutex_lock);
-#endif
     pthread_mutex_destroy(&ctx->mutex_tls);
     pthread_mutex_destroy(&ctx->mutex_thread);
     pthread_mutex_destroy(&ctx->mutex_bridge);
+    pthread_mutex_destroy(&ctx->mutex_lock);
+#endif
 
     freeCycleLog(ctx);
 

src/box86version.h

@@ -2,7 +2,7 @@
 #define __BOX86_VERSION_H_
 
 #define BOX86_MAJOR 0
-#define BOX86_MINOR 2
-#define BOX86_REVISION 8
+#define BOX86_MINOR 3
+#define BOX86_REVISION 1
 
 #endif //__BOX86_VERSION_H_

src/dynarec/arm_emitter.h

@@ -390,6 +390,8 @@ Op is 20-27
 #define LDRD_IMM8(reg, addr, imm8) EMIT(c__ | 0b000<<25 | 1<<24  | (((imm8)<0)?0:1)<<23 | 1<<22 | 0<<21 | 0<<20 | ((reg) << 12) | ((addr) << 16) | ((abs(imm8))&0xf0)<<(8-4) | (0b1101<<4) | ((abs(imm8))&0x0f) )
 // ldrd reg, reg+1, [addr, +rm], reg must be even, reg+1 is implicit
 #define LDRD_REG(reg, addr, rm) EMIT(c__ | 0b000<<25 | 1<<24  | 1<<23 | 0<<22 | 0<<21 | 0<<20 | ((reg) << 12) | ((addr) << 16) | 0b1101<<4 | (rm) )
+// ldr reg, [pc+imm]
+#define LDR_literal(reg, imm12) EMIT(c__ | 0b010<<25 | 1<<24 | (((imm12)>=0)?1:0)<<23 | 0<<21 | 1<<20 | 0b1111<<16 | (reg)<<12 | (abs(imm12)))
 
 // str reg, [addr, #+/-imm9]
 #define STR_IMM9(reg, addr, imm9) EMIT(0xe5000000 | (((imm9)<0)?0:1)<<23 | ((reg) << 12) | ((addr) << 16) | brIMM(imm9) )
@@ -820,6 +822,8 @@ Op is 20-27
 #define VLD1_32(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 15))
 // Load [Rn]! => Dd. Align is 4
 #define VLD1_32_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 13))
+// Load [Rn]! => Dd. Align is 8
+#define VLD1_64_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 3, 0, 13))
 // Load [Rn]! => Dd/Dd+1. Align is 4
 #define VLD1Q_32_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b1010, 2, 0, 13))
 // Load [Rn, Rm]! => Dd/Dd+1. If Rm==15, no writeback, Rm ignored, else writeback Rn <- Rn+Rm. Align is 4
@@ -841,6 +845,8 @@ Op is 20-27
 
 // Store [Rn]! => Dd. Align is 4
 #define VST1_32_W(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 13))
+// Store [Rn]! => Dd. Align is 8
+#define VST1_64_W(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 3, 0, 13))
 // Store [Rn] => Dd/Dd+1. Align is 4
 #define VST1Q_32(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b1010, 2, 0, 15))
 // Store [Rn]! => Dd/Dd+1. Align is 4

src/dynarec/arm_lock_helper.S

@@ -16,7 +16,14 @@
 .global arm_lock_xchg
 .global arm_lock_storeifnull
 .global arm_lock_storeifref
-
+.global arm_lock_storeifref2
+.global arm_lock_decifnot0b
+.global arm_lock_incb
+.global arm_lock_incif0b
+.global arm_lock_stored
+.global arm_lock_storeb
+.global arm_lock_storeif0b
+.global arm_lock_readb
 
 arm_lock_read_b:
     // address is r0, return is r0
@@ -83,33 +90,117 @@ arm_lock_xchg:
     bx      lr
 
 arm_lock_storeifnull:
-    // address is r0, value is r1, r1 store to r0 only if [r0] is 0. return new [r0] value (so r1 or old value)
+    // address is r0, value is r1, r1 store to r0 only if [r0] is 0. return old [r0] value
     // r0 needs to be aligned
     dmb     ish
     ldrex   r2, [r0]
     cmp     r2, #0
     bne     arm_lock_storeifnull_exit
-    mov     r2, r1
-    strex   r3, r2, [r0]
+    strex   r3, r1, [r0]
     cmp     r3, #1
     beq     arm_lock_storeifnull
 arm_lock_storeifnull_exit:
     mov     r0, r2
     bx      lr
 
+arm_lock_storeifref2:
+    dmb     ish
+arm_lock_storeifref2_loop:
+    // address is r0, value is r1, r1 store to r0 only if [r0] is r2. return old [r0] value (so r2 or old value)
+    ldrex   r3, [r0]
+    cmp     r3, r2
+    movne   r0, r3
+    bne     arm_lock_storeifref2_exit
+    strex   r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_storeifref_loop
+    mov     r0, r2
+arm_lock_storeifref2_exit:
+    bx      lr
+
 arm_lock_storeifref:
-    push    {r4}
     dmb     ish
 arm_lock_storeifref_loop:
     // address is r0, value is r1, r1 store to r0 only if [r0] is r2. return new [r0] value (so r1 or old value)
     ldrex   r3, [r0]
     cmp     r3, r2
+    movne   r0, r3
     bne     arm_lock_storeifref_exit
-    mov     r3, r1
-    strex   r4, r3, [r0]
-    cmp     r4, #1
+    strex   r3, r1, [r0]
+    cmp     r3, #1
     beq     arm_lock_storeifref_loop
+    mov     r0, r1
 arm_lock_storeifref_exit:
-    mov     r0, r3
-    pop     {r4}
+    bx      lr
+
+arm_lock_decifnot0b:
+    dmb     ish
+arm_lock_decifnot0b_0:
+    ldrexb  r1, [r0]
+    cmp     r1, #0
+    subne   r3, r1, #1
+    moveq   r3, r1
+    strexb  r2, r3, [r0]
+    cmp     r2, #1
+    beq     arm_lock_decifnot0b_0
+    dmb     ish
+    mov     r0, r1
+    bx      lr
+
+arm_lock_incb:
+    dmb     ish
+arm_lock_incb_0:
+    ldrexb  r1, [r0]
+    add     r1, r1, #1
+    strexb  r2, r1, [r0]
+    cmp     r2, #1
+    beq     arm_lock_incb_0
+arm_lock_incb_exit:
+    dmb     ish
+    bx      lr
+
+arm_lock_incif0b:
+    dmb     ish
+arm_lock_incif0b_0:
+    ldrexb  r1, [r0]
+    cmp     r1, #0
+    addeq   r3, r1, #1
+    movne   r3, r1
+    strexb  r2, r3, [r0]
+    cmp     r2, #1
+    beq     arm_lock_incif0b_0
+    dmb     ish
+    mov     r0, r1
+    bx      lr
+
+arm_lock_stored:
+    dmb     ish
+    str     r1, [r0]
+    dmb     ish
+    bx      lr
+
+arm_lock_storeb:
+    dmb     ish
+    strb    r1, [r0]
+    dmb     ish
+    bx      lr
+
+arm_lock_storeif0b:
+    dmb     ish
+arm_lock_storeif0b_0:
+    ldrexb  r2, [r0]
+    cmp     r2, #0
+    bne     arm_lock_storeif0b_exit
+    strexb  r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_storeif0b_0
+    mov     r2, #0
+    dmb     ish
+arm_lock_storeif0b_exit:
+    mov     r0, r2
+    bx      lr
+
+arm_lock_readb:
+    dmb     ish
+    ldrb    r0, [r0]
     bx      lr

src/dynarec/arm_lock_helper.h

@@ -25,10 +25,34 @@ extern int arm_lock_write_dd(uint32_t a, uint32_t b, void* addr);
 // Atomically exchange value at [p] with val, return old p
 extern uintptr_t arm_lock_xchg(void* p, uintptr_t val);
 
-// Atomically store value to [p] only if [p] is NULL. Return new [p] value (so val or old)
+// Atomically store value to [p] only if [p] is NULL. Return old [p] value
 extern void* arm_lock_storeifnull(void* p, void* val);
 
 // Atomically store value to [p] only if [p] is ref. Return new [p] value (so val or old)
 extern void* arm_lock_storeifref(void* p, void* val, void* ref);
 
+// Atomically store value to [p] only if [p] is ref. Return old [p] value
+extern void* arm_lock_storeifref2(void* p, void* val, void* ref);
+
+// decrement atomicaly the byte at [p] (but only if p not 0). Return old [p] value
+extern int arm_lock_decifnot0b(void*p);
+
+// increment atomicaly the byte at [p]
+extern void arm_lock_incb(void*p);
+
+// increment atomicaly the byte at [p] only if it was 0. Return the old value of [p]
+extern int arm_lock_incif0b(void*p);
+
+// atomic store (with memory barrier)
+extern void arm_lock_stored(void*p, uint32_t d);
+
+// atomic store (with memory barrier)
+extern void arm_lock_storeb(void*p, uint8_t b);
+
+// atomic store (with memory barrier) only if it was 0. Return the old value of [p]
+extern int arm_lock_storeif0b(void*p, uint8_t b);
+
+// atomic load (with memory barrier)
+extern int arm_lock_readb(void*p);
+
 #endif  //__ARM_LOCK_HELPER__H__

src/dynarec/arm_next.S

@@ -9,6 +9,7 @@
 .extern LinkNext
 
 .global arm_next
+    .4byte  0   // NULL pointer before arm64_next, for getDB
 arm_next:
     // emu is r0
     //stm     r0, {r4-r12,r14}   // don't put put back reg value in emu, faster but more tricky to debug

src/dynarec/dynablock.c

@@ -41,21 +41,37 @@ uint32_t X31_hash_code(void* addr, int len)
 	return (uint32_t)h;
 }
 
-dynablocklist_t* NewDynablockList(uintptr_t text, int textsz, int direct)
+dynablock_t* InvalidDynablock(dynablock_t* db, int need_lock)
 {
-    if(!textsz) {
-        printf_log(LOG_NONE, "Error, creating a NULL sized Dynablock\n");
-        return NULL;
+    if(db) {
+        if(db->gone)
+            return NULL; // already in the process of deletion!
+        dynarec_log(LOG_DEBUG, "InvalidDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
+        if(need_lock)
+            mutex_lock(&my_context->mutex_dyndump);
+        // remove jumptable
+        setJumpTableDefault(db->x86_addr);
+        db->done = 0;
+        db->gone = 1;
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
     }
-    dynablocklist_t* ret = (dynablocklist_t*)box_calloc(1, sizeof(dynablocklist_t));
-    ret->text = text;
-    ret->textsz = textsz;
-    if(direct && textsz) {
-        ret->direct = (dynablock_t**)box_calloc(textsz, sizeof(dynablock_t*));
-        if(!ret->direct) {printf_log(LOG_NONE, "Warning, fail to create direct block for dynablock @%p\n", (void*)text);}
+    return db;
+}
+
+void FreeInvalidDynablock(dynablock_t* db, int need_lock)
+{
+    if(db) {
+        if(!db->gone)
+            return; // already in the process of deletion!
+        dynarec_log(LOG_DEBUG, "FreeInvalidDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
+        if(need_lock)
+            mutex_lock(&my_context->mutex_dyndump);
+        FreeDynarecMap((uintptr_t)db->actual_block);
+        customFree(db);
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
     }
-    dynarec_log(LOG_DEBUG, "New Dynablocklist %p, from %p->%p\n", ret, (void*)text, (void*)(text+textsz));
-    return ret;
 }
 
 void FreeDynablock(dynablock_t* db, int need_lock)
@@ -63,239 +79,100 @@ void FreeDynablock(dynablock_t* db, int need_lock)
     if(db) {
         if(db->gone)
             return; // already in the process of deletion!
-        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x86=%p:%p father=%p, with %d son(s) already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size, db->father, db->sons_size, db->gone);
+        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
         if(need_lock)
-            pthread_mutex_lock(&my_context->mutex_dyndump);
-        db->done = 0;
-        db->gone = 1;
-        // remove from direct if there
-        uintptr_t startdb = db->parent->text;
-        uintptr_t enddb = db->parent->text + db->parent->textsz;
-        if(db->parent->direct) {
-            uintptr_t addr = (uintptr_t)db->x86_addr;
-            if(addr>=startdb && addr<enddb)
-                db->parent->direct[addr-startdb] = NULL;
-        }
+            mutex_lock(&my_context->mutex_dyndump);
         // remove jumptable
         setJumpTableDefault(db->x86_addr);
-        // remove and free the sons
-        for (int i=0; i<db->sons_size; ++i) {
-            dynablock_t *son = (dynablock_t*)arm_lock_xchg(&db->sons[i], 0);
-            FreeDynablock(son, 0);
-        }
-        // only the father free the DynarecMap
-        if(!db->father) {
-            dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->block, db->size);
-            FreeDynarecMap(db, (uintptr_t)db->block, db->size);
-            box_free(db->sons);
-            box_free(db->instsize);
-        }
-        box_free(db);
+        dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->actual_block, db->size);
+        db->done = 0;
+        db->gone = 1;
+        if(db->previous)
+            FreeInvalidDynablock(db->previous, 0);
+        FreeDynarecMap((uintptr_t)db->actual_block);
+        customFree(db);
         if(need_lock)
-            pthread_mutex_unlock(&my_context->mutex_dyndump);
+            mutex_unlock(&my_context->mutex_dyndump);
     }
 }
 
-void FreeDynablockList(dynablocklist_t** dynablocks)
-{
-    if(!dynablocks)
-        return;
-    if(!*dynablocks)
-        return;
-    dynarec_log(LOG_DEBUG, "Free Dynablocklist %p, with Direct Blocks %p\n", *dynablocks, (*dynablocks)->direct);
-    if((*dynablocks)->direct) {
-        for (int i=0; i<(*dynablocks)->textsz; ++i) {
-            if((*dynablocks)->direct[i] && !(*dynablocks)->direct[i]->father) 
-                FreeDynablock((*dynablocks)->direct[i], my_context?1:0);
-        }
-        box_free((*dynablocks)->direct);
-    }
-    (*dynablocks)->direct = NULL;
 
-    box_free(*dynablocks);
-    *dynablocks = NULL;
-}
 
 void MarkDynablock(dynablock_t* db)
 {
     if(db) {
-        if(db->father)
-            db = db->father;    // mark only father
-        if(db->need_test)
-            return; // already done
-        db->need_test = 1;
-        setJumpTableDefault(db->x86_addr);
-        for(int i=0; i<db->sons_size; ++i)
-            setJumpTableDefault(db->sons[i]->x86_addr);
+        dynarec_log(LOG_DEBUG, "MarkDynablock %p %p-%p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1);
+        if(!setJumpTableIfRef(db->x86_addr, db->jmpnext, db->block)) {
+            dynablock_t* old = db;
+            db = getDB((uintptr_t)old->x86_addr);
+            if(!old->gone && db!=old) {
+                printf_log(LOG_INFO, "Warning, couldn't mark block as dirty for %p, block=%p, current_block=%p\n", old->x86_addr, old, db);
+                // the block is lost, need to invalidate it...
+                old->gone = 1;
+                old->done = 0;
+                if(!db || db->previous)
+                    FreeInvalidDynablock(old, 1);
+                else
+                    db->previous = old;
+            }
+        }
     }
 }
 
-uintptr_t StartDynablockList(dynablocklist_t* db)
-{
-    if(db)
-        return db->text;
-    return 0;
-}
-uintptr_t EndDynablockList(dynablocklist_t* db)
-{
-    if(db)
-        return db->text+db->textsz-1;
-    return 0;
-}
-
-int IntervalIntersects(uintptr_t start1, uintptr_t end1, uintptr_t start2, uintptr_t end2)
+static int IntervalIntersects(uintptr_t start1, uintptr_t end1, uintptr_t start2, uintptr_t end2)
 {
     if(start1 > end2 || start2 > end1)
         return 0;
     return 1;
 }
 
-void MarkDirectDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
+static int MarkedDynablock(dynablock_t* db)
 {
-    if(!dynablocks)
-        return;
-    if(!dynablocks->direct)
-        return;
-    uintptr_t startdb = dynablocks->text;
-    uintptr_t enddb = startdb + dynablocks->textsz -1;
-    uintptr_t start = addr;
-    uintptr_t end = addr+size-1;
-    if(start<startdb)
-        start = startdb;
-    if(end>enddb)
-        end = enddb;
-    dynablock_t *db;
-    if(end>startdb && start<enddb)
-        for(uintptr_t i = start; i<end; ++i)
-            if((db=dynablocks->direct[i-startdb]))
-                if(IntervalIntersects((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr+db->x86_size-1, addr, addr+size+1))
-                    MarkDynablock(db);
-}
-
-void FreeRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
-{
-    if(!dynablocks)
-        return;
-
-    if(dynablocks->direct) {
-        int need_lock = my_context?1:0;
-        dynablock_t* db;
-        int ret;
-        khint_t k;
-        kh_dynablocks_t *blocks = kh_init(dynablocks);
-        // copy in a temporary list
-        if(dynablocks->direct) {
-            uintptr_t startdb = dynablocks->text;
-            uintptr_t enddb = startdb + dynablocks->textsz;
-            uintptr_t start = addr;
-            uintptr_t end = addr+size;
-            if(start<startdb)
-                start = startdb;
-            if(end>enddb)
-                end = enddb;
-            if(end>startdb && start<enddb)
-                for(uintptr_t i = start; i<end; ++i) {
-                    db = (dynablock_t*)arm_lock_xchg(&dynablocks->direct[i-startdb], 0);
-                    if(db) {
-                        if(db->father)
-                            db = db->father;
-                        if(db->parent==dynablocks) {
-                            k = kh_put(dynablocks, blocks, (uintptr_t)db, &ret);
-                            kh_value(blocks, k) = db;
-                        }
-                    }
-                }
-        }
-        // purge the list
-        kh_foreach_value(blocks, db,
-            FreeDynablock(db, need_lock);
-        );
-        kh_destroy(dynablocks, blocks);
+    if(db) {
+        if(getNeedTest((uintptr_t)db->x86_addr))
+            return 1; // already done
     }
+    return 0;
 }
-void MarkRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
+
+void MarkRangeDynablock(dynablock_t* db, uintptr_t addr, uintptr_t size)
 {
-    if(!dynablocks)
+    // Mark will try to find *any* blocks that intersect the range to mark
+    if(!db)
         return;
-    if(dynablocks->direct) {
-        uintptr_t new_addr = addr - dynablocks->maxsz;
-        uintptr_t new_size = size + dynablocks->maxsz;
-        MarkDirectDynablock(dynablocks, new_addr, new_size);
-        // the blocks check before
-        for(int idx=(new_addr)>>DYNAMAP_SHIFT; idx<(addr>>DYNAMAP_SHIFT); ++idx)
-            MarkDirectDynablock(getDB(idx), new_addr, new_size);
+    dynarec_log(LOG_DEBUG, "MarkRangeDynablock %p-%p .. startdb=%p, sizedb=%p\n", (void*)addr, (void*)addr+size-1, (void*)db->x86_addr, (void*)db->x86_size);
+    if(IntervalIntersects((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr+db->x86_size-1, addr, addr+size+1))
+        MarkDynablock(db);
+}
+
+int FreeRangeDynablock(dynablock_t* db, uintptr_t addr, uintptr_t size)
+{
+    if(!db)
+        return 1;
+
+    int need_lock = my_context?1:0;
+    if(IntervalIntersects((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr+db->x86_size-1, addr, addr+size+1)) {
+        FreeDynablock(db, need_lock);
+        return 0;
     }
+    return 1;
 }
 
-dynablock_t* FindDynablockDynablocklist(void* addr, kh_dynablocks_t* dynablocks)
+dynablock_t *AddNewDynablock(uintptr_t addr)
 {
-    if(!dynablocks)
-        return NULL;
-    dynablock_t* db;
-    kh_foreach_value(dynablocks, db, 
-        const uintptr_t s = (uintptr_t)db->block;
-        const uintptr_t e = (uintptr_t)db->block+db->size;
-        if((uintptr_t)addr>=s && (uintptr_t)addr<e)
-            return db->father?db->father:db;
-    )
-    return NULL;
-}
-
-static dynablocklist_t* getDBFromAddress(uintptr_t addr)
-{
-    const int idx = (addr>>DYNAMAP_SHIFT);
-    return getDB(idx);
-}
-
-dynablock_t *AddNewDynablock(dynablocklist_t* dynablocks, uintptr_t addr, int* created)
-{
-    if(!dynablocks) {
-        dynarec_log(LOG_INFO, "Warning: Ask to create a dynablock with a NULL dynablocklist (addr=%p)\n", (void*)addr);
-        *created = 0;
+    dynablock_t* block;
+    #if 0
+    // check if memory as the correct flags
+    int prot = getProtection(addr);
+    if(!(prot&(PROT_EXEC|PROT_DYNAREC|PROT_DYNAREC_R))) {
+        dynarec_log(LOG_VERBOSE, "Block asked on a memory with no execution flags 0x%02X\n", prot);
         return NULL;
     }
-    if((addr<dynablocks->text) || (addr>=(dynablocks->text+dynablocks->textsz))) {
-        // this should be useless
-        //dynarec_log(LOG_INFO, "Warning: Refused to create a Direct Block that is out-of-bound: dynablocks=%p (%p:%p), addr=%p\n", dynablocks, (void*)(dynablocks->text), (void*)(dynablocks->text+dynablocks->textsz), (void*)addr);
-        //*created = 0;
-        //return NULL;
-        return AddNewDynablock(getDBFromAddress(addr), addr, created);
-    }
-    dynablock_t* block = NULL;
-    // first, check if it exist in direct access mode
-    if(dynablocks->direct) {
-        block = dynablocks->direct[addr-dynablocks->text];
-        if(block) {
-            dynarec_log(LOG_VERBOSE, "Block already exist in Direct Map\n");
-            *created = 0;
-            return block;
-        }
-    }
     
-    if (!*created)
-        return block;
-    
-    pthread_mutex_lock(&my_context->mutex_dyndump);
-    if(!dynablocks->direct) {
-        dynablock_t** p = (dynablock_t**)box_calloc(dynablocks->textsz, sizeof(dynablock_t*));
-        if(arm_lock_storeifnull(&dynablocks->direct, p)!=p)
-            box_free(p);    // someone already create the direct array, too late...
-    }
-
+    #endif
     // create and add new block
     dynarec_log(LOG_VERBOSE, "Ask for DynaRec Block creation @%p\n", (void*)addr);
-
-    block = (dynablock_t*)box_calloc(1, sizeof(dynablock_t));
-    block->parent = dynablocks; 
-    dynablock_t* tmp = (dynablock_t*)arm_lock_storeifnull(&dynablocks->direct[addr-dynablocks->text], block);
-    pthread_mutex_unlock(&my_context->mutex_dyndump);
-    if(tmp !=  block) {
-        // a block appeard!
-        box_free(block);
-        *created = 0;
-        return tmp;
-    }
-    *created = 1;
+    block = (dynablock_t*)customCalloc(1, sizeof(dynablock_t));
     return block;
 }
 
@@ -311,114 +188,120 @@ void cancelFillBlock()
     return NULL if block is not found / cannot be created. 
     Don't create if create==0
 */
-static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, dynablock_t* current, int need_lock)
+static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, int need_lock)
 {
     if(hasAlternate((void*)addr))
         return NULL;
-    // try the quickest way first: get parent of current and check if ok!
-    dynablocklist_t *dynablocks = NULL;
-    dynablock_t* block = NULL;
-    if(current && current->done) {
-        dynablocks = current->parent;
-        if(dynablocks && !(addr>=dynablocks->text && addr<(dynablocks->text+dynablocks->textsz)))
-            dynablocks = NULL;
-    }
-    // nope, lets do the long way
-    if(!dynablocks) {
-        dynablocks = getDBFromAddress(addr);
-        if(!dynablocks) {
-            dynablocks = GetDynablocksFromAddress(emu->context, addr);
-            if(!dynablocks)
+    dynablock_t* block = getDB(addr);
+    if(block || !create)
+        return block;
+
+    if(need_lock) {
+        if(box86_dynarec_wait) {
+            mutex_lock(&my_context->mutex_dyndump);
+        } else {
+            if(mutex_trylock(&my_context->mutex_dyndump))   // FillBlock not available for now
                 return NULL;
         }
-    }
-    // check direct first, without lock
-    if(dynablocks->direct/* && (addr>=dynablocks->text) && (addr<(dynablocks->text+dynablocks->textsz))*/)
-        if((block = dynablocks->direct[addr-dynablocks->text]))
+        block = getDB(addr);    // just in case
+        if(block) {
+            mutex_unlock(&my_context->mutex_dyndump);
             return block;
-
-    int created = create;
-    block = AddNewDynablock(dynablocks, addr, &created);
-    if(!created)
-        return block;   // existing block...
+        }
+    }
+    
+    block = AddNewDynablock(addr);
 
     // fill the block
     block->x86_addr = (void*)addr;
-    if(need_lock)
-        pthread_mutex_lock(&my_context->mutex_dyndump);
     if(sigsetjmp(&dynarec_jmpbuf, 1)) {
         printf_log(LOG_INFO, "FillBlock at %p triggered a segfault, cancelling\n", (void*)addr);
+        FreeDynablock(block, 0);
         if(need_lock)
-            pthread_mutex_unlock(&my_context->mutex_dyndump);
+            mutex_unlock(&my_context->mutex_dyndump);
         return NULL;
     }
     void* ret = FillBlock(block, filladdr);
-    if(need_lock)
-        pthread_mutex_unlock(&my_context->mutex_dyndump);
     if(!ret) {
         dynarec_log(LOG_DEBUG, "Fillblock of block %p for %p returned an error\n", block, (void*)addr);
-        void* old = (void*)arm_lock_storeifref(&dynablocks->direct[addr-dynablocks->text], 0, block);
-        if(old!=block && old) { // put it back in place, strange things are happening here!
-            dynarec_log(LOG_INFO, "Warning, a wild block appeared at %p: %p\n", (void*)addr, old);
-            // Doing nothing else, the block has not been written
-        }
-        box_free(block);
+        customFree(block);
         block = NULL;
     }
     // check size
-    if(block && block->x86_size) {
+    if(block) {
         int blocksz = block->x86_size;
-        if(dynablocks->maxsz<blocksz) {
-            dynablocks->maxsz = blocksz;
-            for(int idx=(addr>>DYNAMAP_SHIFT)+1; idx<=((addr+blocksz)>>DYNAMAP_SHIFT); ++idx) {
-                dynablocklist_t* dblist;
-                if((dblist = getDB(idx)))
-                    if(dblist->maxsz<blocksz)
-                        dblist->maxsz = blocksz;
-            }
-        }
-        //protectDB((uintptr_t)block->x86_addr, block->x86_size);   // already protected in FillBlock
-        // but check if it already need test
-        if(block->need_test || (block->father && block->father->need_test))
-            return NULL;
+        if(blocksz>my_context->max_db_size)
+            my_context->max_db_size = blocksz;
         // fill-in jumptable
-        addJumpTableIfDefault(block->x86_addr, block->block);
-        for(int i=0; i<block->sons_size; ++i)
-            addJumpTableIfDefault(block->sons[i]->x86_addr, block->sons[i]->block);
+        if(!addJumpTableIfDefault(block->x86_addr, block->dirty?block->jmpnext:block->block)) {
+            FreeDynablock(block, 0);
+            block = getDB(addr);
+            MarkDynablock(block);   // just in case...
+        } else {
+            if(block->x86_size)
+                block->done = 1;    // don't validate the block if the size is null, but keep the block
+        }
     }
+    if(need_lock)
+        mutex_unlock(&my_context->mutex_dyndump);
 
-    dynarec_log(LOG_DEBUG, " --- DynaRec Block %s @%p:%p (%p, 0x%x bytes, with %d son(s))\n", created?"created":"recycled", (void*)addr, (void*)(addr+((block)?block->x86_size:0)), (block)?block->block:0, (block)?block->size:0, (block)?block->sons_size:0);
+    dynarec_log(LOG_DEBUG, "%04d| --- DynaRec Block created @%p:%p (%p, 0x%x bytes)\n", GetTID(), (void*)addr, (void*)(addr+((block)?block->x86_size:1)-1), (block)?block->block:0, (block)?block->size:0);
 
     return block;
 }
 
-dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create, dynablock_t** current)
+dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create)
 {
-    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, *current, 1);
-    if(db && db->done && db->block && (db->need_test || (db->father && db->father->need_test))) {
-        if(pthread_mutex_trylock(&my_context->mutex_dyndump))
-            return NULL;
-        dynablock_t *father = db->father?db->father:db;
-        uint32_t hash = (getProtection((uintptr_t)father->x86_addr)&PROT_READ)?X31_hash_code(father->x86_addr, father->x86_size):0;
-        if(hash!=father->hash) {
-            dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) with %d son(s) for %p\n", father, father->x86_addr, father->x86_addr+father->x86_size, hash, father->hash, father->sons_size, (void*)addr);
-            // no more current if it gets invalidated too
-            if(*current && father->x86_addr>=(*current)->x86_addr && (father->x86_addr+father->x86_size)<(*current)->x86_addr)
-                *current = NULL;
-            // Free father, it's now invalid!
-            FreeDynablock(father, 0);
-            // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, addr, create, *current, 0);
-        } else {
-            father->need_test = 0;
-            protectDB((uintptr_t)father->x86_addr, father->x86_size);
-            // fill back jumptable
-            addJumpTableIfDefault(father->x86_addr, father->block);
-            for(int i=0; i<father->sons_size; ++i)
-                addJumpTableIfDefault(father->sons[i]->x86_addr, father->sons[i]->block);
+    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, 1);
+    if(db && db->done && db->block && getNeedTest(addr)) {
+        if(db->always_test)
+            sched_yield();  // just calm down...
+        if(AreaInHotPage((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr + db->x86_size - 1)) {
+            emu->test.test = 0;
+            if(box86_dynarec_fastpage) {
+                uint32_t hash = X31_hash_code(db->x86_addr, db->x86_size);
+                if(hash==db->hash) { // seems ok, run it without reprotecting it
+                    setJumpTableIfRef(db->x86_addr, db->block, db->jmpnext);
+                    return db;
+                }
+                db->done = 0;   // invalidating the block, it's already not good
+                dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, hash, db->hash, (void*)addr);
+                // Free db, it's now invalid!
+                FreeDynablock(db, 1);
+                return NULL;    // not building a new one, it's still a hotpage
+            } else {
+                dynarec_log(LOG_INFO, "Not running block %p from %p:%p with for %p because it's in a hotpage\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, (void*)addr);
+                return NULL;
+            }
         }
-        pthread_mutex_unlock(&my_context->mutex_dyndump);
+        uint32_t hash = X31_hash_code(db->x86_addr, db->x86_size);
+        int need_lock = mutex_trylock(&my_context->mutex_dyndump);
+        if(hash!=db->hash) {
+            db->done = 0;   // invalidating the block
+            dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, hash, db->hash, (void*)addr);
+            // Free db, it's now invalid!
+            dynablock_t* old = InvalidDynablock(db, need_lock);
+            // start again... (will create a new block)
+            db = internalDBGetBlock(emu, addr, addr, create, need_lock);
+            if(db) {
+                if(db->previous)
+                    FreeInvalidDynablock(db->previous, need_lock);
+                db->previous = old;
+            } else
+                FreeInvalidDynablock(old, need_lock);
+        } else {
+            dynarec_log(LOG_DEBUG, "Validating block %p from %p:%p (hash:%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, db->hash, (void*)addr);
+            protectDB((uintptr_t)db->x86_addr, db->x86_size);
+            // fill back jumptable
+            if(isprotectedDB((uintptr_t)db->x86_addr, db->x86_size) && !db->always_test) {
+                setJumpTableIfRef(db->x86_addr, db->block, db->jmpnext);
+            }
+        }
+        if(!need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
     } 
+    if(!db || !db->block || !db->done)
+        emu->test.test = 0;
     return db;
 }
 
@@ -426,27 +309,36 @@ dynablock_t* DBAlternateBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr)
 {
     dynarec_log(LOG_DEBUG, "Creating AlternateBlock at %p for %p\n", (void*)addr, (void*)filladdr);
     int create = 1;
-    dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, NULL, 1);
-    if(db && db->done && db->block && (db->need_test || (db->father && db->father->need_test))) {
-        if(pthread_mutex_trylock(&my_context->mutex_dyndump))
-            return NULL;
-        dynablock_t *father = db->father?db->father:db;
-        uint32_t hash = (getProtection((uintptr_t)father->x86_addr)&PROT_READ)?X31_hash_code(father->x86_addr, father->x86_size):0;
-        if(hash!=father->hash) {
-            dynarec_log(LOG_DEBUG, "Invalidating alt block %p from %p:%p (hash:%X/%X) with %d son(s) for %p\n", father, father->x86_addr, father->x86_addr+father->x86_size, hash, father->hash, father->sons_size, (void*)addr);
-            // Free father, it's now invalid!
-            FreeDynablock(father, 0);
+    dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, 1);
+    if(db && db->done && db->block && getNeedTest(filladdr)) {
+        if(db->always_test)
+            sched_yield();  // just calm down...
+        int need_lock = mutex_trylock(&my_context->mutex_dyndump);
+        uint32_t hash = X31_hash_code(db->x86_addr, db->x86_size);
+        if(hash!=db->hash) {
+            db->done = 0;   // invalidating the block
+            dynarec_log(LOG_DEBUG, "Invalidating alt block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size, hash, db->hash, (void*)addr);
+            // Free db, it's now invalid!
+            dynablock_t* old = InvalidDynablock(db, need_lock);
             // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, filladdr, create, NULL, 0);
+            db = internalDBGetBlock(emu, addr, filladdr, create, need_lock);
+            if(db) {
+                if(db->previous)
+                    FreeInvalidDynablock(db->previous, need_lock);
+                db->previous = old;
+            } else
+                FreeInvalidDynablock(old, need_lock);
         } else {
-            father->need_test = 0;
-            protectDB((uintptr_t)father->x86_addr, father->x86_size);
+            protectDB((uintptr_t)db->x86_addr, db->x86_size);
             // fill back jumptable
-            addJumpTableIfDefault(father->x86_addr, father->block);
-            for(int i=0; i<father->sons_size; ++i)
-                addJumpTableIfDefault(father->sons[i]->x86_addr, father->sons[i]->block);
+            if(isprotectedDB((uintptr_t)db->x86_addr, db->x86_size) && !db->always_test) {
+                setJumpTableIfRef(db->x86_addr, db->block, db->jmpnext);
+            }
         }
-        pthread_mutex_unlock(&my_context->mutex_dyndump);
+        if(!need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
     } 
+    if(!db || !db->block || !db->done)
+        emu->test.test = 0;
     return db;
 }

src/dynarec/dynablock_private.h

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

src/dynarec/dynarec.c

@@ -20,6 +20,7 @@
 #include "dynablock.h"
 #include "dynablock_private.h"
 #include "bridge.h"
+void x86test_check(x86emu_t* ref, uintptr_t ip);
 #endif
 
 #ifdef ARM
@@ -44,9 +45,8 @@ void* LinkNext(x86emu_t* emu, uintptr_t addr, void* x2)
         printf_log(LOG_NONE, "Warning, jumping to NULL address from %p (db=%p, x86addr=%p/%s)\n", x2, db, x86addr, pcname);
     }
     #endif
-    dynablock_t* current = NULL;
     void * jblock;
-    dynablock_t* block = DBGetBlock(emu, addr, 1, &current);
+    dynablock_t* block = DBGetBlock(emu, addr, 1);
     if(!block) {
         // no block, let link table as is...
         if(hasAlternate((void*)addr)) {
@@ -54,12 +54,15 @@ void* LinkNext(x86emu_t* emu, uintptr_t addr, void* x2)
             addr = (uintptr_t)getAlternate((void*)addr);
             R_EIP = addr;
             printf_log(LOG_INFO, " -> %p\n", (void*)addr);
-            block = DBGetBlock(emu, addr, 1, &current);
+            block = DBGetBlock(emu, addr, 1);
         }
         if(!block) {
             #ifdef HAVE_TRACE
-            dynablock_t* db = FindDynablockFromNativeAddress(x2);
-            dynarec_log(LOG_INFO, "Warning, jumping to a no-block address %p from %p (db=%p, x86addr=%p)\n", (void*)addr, x2, db, db?(void*)getX86Address(db, (uintptr_t)x2):NULL);
+            if(LOG_INFO<=box86_dynarec_log) {
+                dynablock_t* db = FindDynablockFromNativeAddress(x2);
+                elfheader_t* h = FindElfAddress(my_context, (uintptr_t)x2);
+                dynarec_log(LOG_INFO, "Warning, jumping to a no-block address %p from %p (db=%p, x64addr=%p(elf=%s))\n", (void*)addr, x2, db, db?(void*)getX86Address(db, (uintptr_t)x2-4):NULL, h?ElfName(h):"(none)");
+            }
             #endif
             //tableupdate(arm_epilog, addr, table);
             return arm_epilog;
@@ -84,18 +87,25 @@ void DynaCall(x86emu_t* emu, uintptr_t addr)
     // prepare setjump for signal handling
     emu_jmpbuf_t *ejb = NULL;
     int jmpbuf_reset = 0;
+    int skip = 0;
     if(emu->type == EMUTYPE_MAIN) {
         ejb = GetJmpBuf();
         if(!ejb->jmpbuf_ok) {
             ejb->emu = emu;
             ejb->jmpbuf_ok = 1;
             jmpbuf_reset = 1;
-            int a;
-            if((a=sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))) {
-                printf_log(LOG_DEBUG, "Setjmp DynaCall %d, fs=0x%x\n", a, ejb->emu->segs[_FS]);
+            if((skip=sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))) {
+                printf_log(LOG_DEBUG, "Setjmp DynaCall %d, fs=0x%x\n", skip, ejb->emu->segs[_FS]);
                 addr = R_EIP;   // not sure if it should still be inside DynaCall!
-                if(a==2)
-                    Run(emu, 1);    // "single step" next instruction that is doing auto-smc
+                #ifdef DYNAREC
+                if(box86_dynarec_test) {
+                    if(emu->test.clean)
+                        x86test_check(emu, R_EIP);
+                    emu->test.clean = 0;
+                }
+                #endif
+                if(skip!=2)
+                    skip = 0;
             }
         }
     }
@@ -117,15 +127,18 @@ void DynaCall(x86emu_t* emu, uintptr_t addr)
         dynablock_t* block = NULL;
         dynablock_t* current = NULL;
         while(!emu->quit) {
-            block = DBGetBlock(emu, R_EIP, 1, &current);
+            block = (skip==2)?NULL:DBGetBlock(emu, R_EIP, 1);
             current = block;
             if(!block || !block->block || !block->done) {
+                skip = 0;
                 // no block, of block doesn't have DynaRec content (yet, temp is not null)
                 // Use interpreter (should use single instruction step...)
                 dynarec_log(LOG_DEBUG, "%04d|Calling Interpretor @%p, emu=%p\n", GetTID(), (void*)R_EIP, emu);
+                if(box86_dynarec_test)
+                    emu->test.clean = 0;
                 Run(emu, 1);
             } else {
-                dynarec_log(LOG_DEBUG, "%04d|Calling DynaRec Block @%p (%p) of %d x86 instructions (father=%p) emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize ,block->father, emu);
+                dynarec_log(LOG_DEBUG, "%04d|Calling DynaRec Block @%p (%p) of %d x86 instructions emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize ,emu);
                 CHECK_FLAGS(emu);
                 // block is here, let's run it!
                 #ifdef ARM
@@ -173,6 +186,7 @@ int DynaRun(x86emu_t* emu)
 {
     // prepare setjump for signal handling
     emu_jmpbuf_t *ejb = NULL;
+    int skip = 0;
 #ifdef DYNAREC
     int jmpbuf_reset = 1;
 #endif
@@ -185,11 +199,17 @@ int DynaRun(x86emu_t* emu)
 #ifdef DYNAREC
             jmpbuf_reset = 1;
 #endif
-            if((a=sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))) {
-                printf_log(LOG_DEBUG, "Setjmp DynaRun %d, fs=0x%x\n", a, ejb->emu->segs[_FS]);
-                if(a==2)
-                    Run(emu, 1);    // "single step" next instruction that is doing auto-smc
-
+            if((skip=sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))) {
+                printf_log(LOG_DEBUG, "Setjmp DynaRun %d, fs=0x%x\n", skip, ejb->emu->segs[_FS]);
+                #ifdef DYNAREC
+                if(box86_dynarec_test) {
+                    if(emu->test.clean)
+                        x86test_check(emu, R_EIP);
+                    emu->test.clean = 0;
+                }
+                #endif
+                if(skip!=2)
+                    skip = 0;
             }
         }
     }
@@ -202,15 +222,18 @@ int DynaRun(x86emu_t* emu)
         dynablock_t* block = NULL;
         dynablock_t* current = NULL;
         while(!emu->quit) {
-            block = DBGetBlock(emu, R_EIP, 1, &current);
+            block = (skip==2)?NULL:DBGetBlock(emu, R_EIP, 1);
             current = block;
             if(!block || !block->block || !block->done) {
+                skip = 0;
                 // no block, of block doesn't have DynaRec content (yet, temp is not null)
                 // Use interpreter (should use single instruction step...)
                 dynarec_log(LOG_DEBUG, "%04d|Running Interpretor @%p, emu=%p\n", GetTID(), (void*)R_EIP, emu);
+                if(box86_dynarec_test)
+                    emu->test.clean = 0;
                 Run(emu, 1);
             } else {
-                dynarec_log(LOG_DEBUG, "%04d|Running DynaRec Block @%p (%p) of %d x86 insts (father=%p) emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize, block->father, emu);
+                dynarec_log(LOG_DEBUG, "%04d|Running DynaRec Block @%p (%p) of %d x86 insts emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize, emu);
                 // block is here, let's run it!
                 #ifdef ARM
                 arm_prolog(emu, block->block);

src/dynarec/dynarec_arm.c

@@ -73,8 +73,8 @@ void add_next(dynarec_arm_t *dyn, uintptr_t addr) {
         }
     // add slots
     if(dyn->next_sz == dyn->next_cap) {
-        dyn->next_cap += 16;
-        dyn->next = (uintptr_t*)box_realloc(dyn->next, dyn->next_cap*sizeof(uintptr_t));
+        dyn->next_cap += 64;
+        dyn->next = (uintptr_t*)customRealloc(dyn->next, dyn->next_cap*sizeof(uintptr_t));
     }
     dyn->next[dyn->next_sz++] = addr;
 }
@@ -83,11 +83,13 @@ uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
     uintptr_t best = 0;
     int i = 0;
     while((i<dyn->next_sz) && (best!=addr)) {
-        if(dyn->next[i]<addr) { // remove the address, it's before current address
-            dyn->next[i] = 0;
-        } else {
-            if((dyn->next[i]<best) || !best)
-                best = dyn->next[i];
+        if(dyn->next[i]) {
+            if(dyn->next[i]<addr) { // remove the address, it's before current address
+                dyn->next[i] = 0;
+            } else {
+                if((dyn->next[i]<best) || !best)
+                    best = dyn->next[i];
+            }
         }
         ++i;
     }
@@ -241,35 +243,31 @@ int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n)
     return (i==n)?1:0;
 }
 
-static instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x86_size, int arm_size)
+void addInst(instsize_t* insts, size_t* size, int x86_size, int native_size)
 {
     // x86 instruction is <16 bytes
     int toadd;
-    if(x86_size>arm_size)
+    if(x86_size>native_size)
         toadd = 1 + x86_size/15;
     else
-        toadd = 1 + arm_size/15;
-    if((*size)+toadd>(*cap)) {
-        *cap = (*size)+toadd;
-        insts = (instsize_t*)box_realloc(insts, (*cap)*sizeof(instsize_t));
-    }
+        toadd = 1 + native_size/15;
     while(toadd) {
         if(x86_size>15)
             insts[*size].x86 = 15;    
         else
             insts[*size].x86 = x86_size;
         x86_size -= insts[*size].x86;
-        if(arm_size>15)
+        if(native_size>15)
             insts[*size].nat = 15;
         else
-            insts[*size].nat = arm_size;
-        arm_size -= insts[*size].nat;
+            insts[*size].nat = native_size;
+        native_size -= insts[*size].nat;
         ++(*size);
         --toadd;
     }
-    return insts;
 }
 
+
 static void fillPredecessors(dynarec_arm_t* dyn)
 {
     int pred_sz = 1;    // to be safe
@@ -288,7 +286,7 @@ static void fillPredecessors(dynarec_arm_t* dyn)
             ++dyn->insts[i+1].pred_sz;
         }
     }
-    dyn->predecessor = (int*)box_malloc(pred_sz*sizeof(int));
+    dyn->predecessor = (int*)customMalloc(pred_sz*sizeof(int));
     // fill pred pointer
     int* p = dyn->predecessor;
     for(int i=0; i<dyn->size; ++i) {
@@ -298,12 +296,15 @@ static void fillPredecessors(dynarec_arm_t* dyn)
     }
     // fill pred
     for(int i=0; i<dyn->size; ++i) {
-        if(i!=dyn->size-1 && dyn->insts[i].x86.has_next && (!i || dyn->insts[i].pred_sz))
+        if((i!=dyn->size-1) && dyn->insts[i].x86.has_next)
             dyn->insts[i+1].pred[dyn->insts[i+1].pred_sz++] = i;
-        if(dyn->insts[i].x86.jmp && dyn->insts[i].x86.jmp_insts!=-1)
-            dyn->insts[dyn->insts[i].x86.jmp_insts].pred[dyn->insts[dyn->insts[i].x86.jmp_insts].pred_sz++] = i;
+        if(dyn->insts[i].x86.jmp && (dyn->insts[i].x86.jmp_insts!=-1)) {
+            int j = dyn->insts[i].x86.jmp_insts;
+            dyn->insts[j].pred[dyn->insts[j].pred_sz++] = i;
+        }
     }
 
+
 }
 
 // updateNeed goes backward, from last intruction to top
@@ -353,36 +354,85 @@ static int updateNeed(dynarec_arm_t* dyn, int ninst, uint8_t need) {
     return ninst;
 }
 
+void* current_helper = NULL;
+
+void CancelBlock(int need_lock)
+{
+    if(need_lock)
+        mutex_lock(&my_context->mutex_dyndump);
+    dynarec_arm_t* helper = (dynarec_arm_t*)current_helper;
+    current_helper = NULL;
+    if(helper) {
+        customFree(helper->next);
+        customFree(helper->insts);
+        customFree(helper->predecessor);
+        if(helper->dynablock && helper->dynablock->actual_block) {
+            FreeDynarecMap((uintptr_t)helper->dynablock->actual_block);
+            helper->dynablock->actual_block = NULL;
+        }
+    }
+    if(need_lock)
+        mutex_unlock(&my_context->mutex_dyndump);
+}
+
 uintptr_t arm_pass0(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass1(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass2(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass3(dynarec_arm_t* dyn, uintptr_t addr);
+void arm_epilog();
+void arm_next();
 
-__thread void* current_helper = NULL;
-
-void CancelBlock()
-{
-    dynarec_arm_t* helper = (dynarec_arm_t*)current_helper;
-    current_helper = NULL;
-    if(!helper)
-        return;
-    box_free(helper->next);
-    box_free(helper->insts);
-    box_free(helper->predecessor);
-    if(helper->dynablock && helper->dynablock->block)
-        FreeDynarecMap(helper->dynablock, (uintptr_t)helper->dynablock->block, helper->dynablock->size);
+void* CreateEmptyBlock(dynablock_t* block, uintptr_t addr) {
+    block->isize = 0;
+    block->done = 0;
+    size_t sz = 8*sizeof(void*);
+    void* actual_p = (void*)AllocDynarecMap(sz);
+    void* p = actual_p + sizeof(void*);
+    if(actual_p==NULL) {
+        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, cancelling block\n", block, sz);
+        CancelBlock(0);
+        return NULL;
+    }
+    block->size = sz;
+    block->actual_block = actual_p;
+    block->block = p;
+    block->jmpnext = p;
+    *(dynablock_t**)actual_p = block;
+    *(void**)(p+4*sizeof(void*)) = arm_epilog;
+    CreateJmpNext(block->jmpnext, p+4*sizeof(void*));
+    // all done...
+    __clear_cache(actual_p, actual_p+sz);   // need to clear the cache before execution...
+    return block;
 }
 
 void* FillBlock(dynablock_t* block, uintptr_t addr) {
+    /*
+        A Block must have this layout:
+
+        0x0000..0x0003  : dynablock_t* : self
+        0x0004..4+4*n   : actual Native instructions, (n is the total number)
+        B ..    B+3     : dynablock_t* : self (as part of JmpNext, that simulate another block)
+        B+4 ..  B+15    : 3 Native code for jmpnext (or jmp epilog in case of empty block)
+        B+16 .. B+19    : jmpnext (or jmp_epilog) address
+        B+20 .. B+31    : empty (in case an architecture needs more than 3 opcodes)
+        B+32 .. B+32+sz : instsize (compressed array with each instruction lenght on x86 and native side)
+
+    */
 dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
+    if(IsInHotPage(addr)) {
+        dynarec_log(LOG_DEBUG, "Cancelling dynarec FillBlock on hotpage for %p\n", (void*)addr);
+        return NULL;
+    }
     if(addr>=box86_nodynarec_start && addr<box86_nodynarec_end) {
         dynarec_log(LOG_DEBUG, "Asked to fill a block in fobidden zone\n");
+        return CreateEmptyBlock(block, addr);
+    }
+    if(current_helper) {
+        dynarec_log(LOG_DEBUG, "Cancelling dynarec FillBlock at %p as anothor one is going on\n", (void*)addr);
         return NULL;
     }
-    if(!isJumpTableDefault((void*)addr)) {
-        dynarec_log(LOG_DEBUG, "Asked to fill a block at %p, but JumpTable is not default\n", (void*)addr);
-        return NULL;
-    }
+    // protect the 1st page
+    protectDB(addr, 1);
     // init the helper
     dynarec_arm_t helper = {0};
     current_helper = &helper;
@@ -390,27 +440,37 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
     helper.start = addr;
     uintptr_t start = addr;
     helper.cap = 64; // needs epilog handling
-    helper.insts = (instruction_arm_t*)box_calloc(helper.cap, sizeof(instruction_arm_t));
+    helper.insts = (instruction_arm_t*)customCalloc(helper.cap, sizeof(instruction_arm_t));
     // pass 0, addresses, x86 jump addresses, overall size of the block
     uintptr_t end = arm_pass0(&helper, addr);
     // no need for next anymore
-    box_free(helper.next);
+    customFree(helper.next);
     helper.next_sz = helper.next_cap = 0;
     helper.next = NULL;
     // basic checks
     if(!helper.size) {
         dynarec_log(LOG_DEBUG, "Warning, null-sized dynarec block (%p)\n", (void*)addr);
-        CancelBlock();
-        return (void*)block;
+        CancelBlock(0);
+        return CreateEmptyBlock(block, addr);
+    }
+    if(!isprotectedDB(addr, 1)) {
+        dynarec_log(LOG_INFO, "Warning, write on current page on pass0, aborting dynablock creation (%p)\n", (void*)addr);
+        AddHotPage(addr);
+        CancelBlock(0);
+        return NULL;
     }
     // already protect the block and compute hash signature
-    protectDB(addr, end-addr);  //end is 1byte after actual end
+    // protect the block of it goes over the 1st page
+    if((addr&~box86_pagesize)!=(end&~box86_pagesize)) // need to protect some other pages too
+        protectDB(addr, end-addr);  //end is 1byte after actual end
     uint32_t hash = X31_hash_code((void*)addr, end-addr);
     // calculate barriers
     for(int i=0; i<helper.size; ++i)
         if(helper.insts[i].x86.jmp) {
             uintptr_t j = helper.insts[i].x86.jmp;
-            if(j<start || j>=end) {
+            if(j<start || j>=end || j==helper.insts[i].x86.addr) {
+                if(j==helper.insts[i].x86.addr) // if there is a loop on some opcode, make the block "always to tested"
+                    helper.always_test = 1;
                 helper.insts[i].x86.jmp_insts = -1;
                 helper.insts[i].x86.need_after |= X_PEND;
             } else {
@@ -427,44 +487,7 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
         }
     // fill predecessors with the jump address
     fillPredecessors(&helper);
-    // check for the optionnal barriers now
-    /*for(int i=helper.size-1; i>=0; --i) {
-        if(helper.insts[i].barrier_maybe) {
-            // out-of-block jump
-            if(helper.insts[i].x86.jmp_insts == -1) {
-                // nothing for now
-            } else {
-                // inside block jump
-                int k = helper.insts[i].x86.jmp_insts;
-                if(k>i) {
-                    // jump in the future
-                    if(helper.insts[k].pred_sz>1) {
-                        // with multiple flow, put a barrier
-                        helper.insts[k].x86.barrier|=BARRIER_FLAGS;
-                    }
-                } else {
-                    // jump back
-                    helper.insts[k].x86.barrier|=BARRIER_FLAGS;
-                }
-            }
-	}
-    }*/
-    // check to remove useless barrier, in case of jump when destination doesn't needs flags
-    /*for(int i=helper.size-1; i>=0; --i) {
-        int k;
-        if(helper.insts[i].x86.jmp
-        && ((k=helper.insts[i].x86.jmp_insts)>=0)
-        && helper.insts[k].x86.barrier&BARRIER_FLAGS) {
-            //TODO: optimize FPU barrier too
-            if((!helper.insts[k].x86.need_flags)
-             ||(helper.insts[k].x86.set_flags==X_ALL
-                  && helper.insts[k].x86.state_flags==SF_SET)
-             ||(helper.insts[k].x86.state_flags==SF_SET_PENDING)) {
-                //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Removed barrier for inst %d\n", k);
-                helper.insts[k].x86.barrier &= ~BARRIER_FLAGS; // remove flag barrier
-             }
-        }
-    }*/
+
     int pos = helper.size;
     while (pos>=0)
         pos = updateNeed(&helper, pos, 0);
@@ -475,29 +498,68 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
     // pass 2, instruction size
     arm_pass2(&helper, addr);
     // ok, now allocate mapped memory, with executable flag on
-    int sz = helper.arm_size;
-    void* p = (void*)AllocDynarecMap(block, sz);
-    if(p==NULL) {
-        dynarec_log(LOG_DEBUG, "AllocDynarecMap(%p, %d) failed, cancelling block\n", block, sz);
-        CancelBlock();
+    size_t insts_rsize = (helper.insts_size+2)*sizeof(instsize_t);
+    insts_rsize = (insts_rsize+7)&~7;   // round the size...
+    size_t arm_size = (helper.arm_size+7)&~7;   // round the size...
+    // ok, now allocate mapped memory, with executable flag on
+    size_t sz = sizeof(void*) + arm_size + 8*sizeof(void*) + insts_rsize;
+    //           dynablock_t* block (arm insts) jmpnext code    instsize
+    void* actual_p = (void*)AllocDynarecMap(sz);
+    void* p = (void*)(((uintptr_t)actual_p) + sizeof(void*));
+    void* next = p + arm_size;
+    void* instsize = next + 8*sizeof(void*);
+    if(actual_p==NULL) {
+        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, cancelling block\n", block, sz);
+        CancelBlock(0);
         return NULL;
     }
     helper.block = p;
+    block->actual_block = actual_p;
     helper.arm_start = (uintptr_t)p;
-    if(helper.sons_size) {
-        helper.sons_x86 = (uintptr_t*)alloca(helper.sons_size*sizeof(uintptr_t));
-        helper.sons_arm = (void**)alloca(helper.sons_size*sizeof(void*));
-    }
-    // pass 3, emit (log emit arm opcode)
+    helper.jmp_next = (uintptr_t)next+sizeof(void*);
+    helper.insts_size = 0;  // reset
+    helper.instsize = (instsize_t*)instsize;
+    *(dynablock_t**)actual_p = block;
+    // pass 3, emit (log emit native opcode)
     if(box86_dynarec_dump) {
         dynarec_log(LOG_NONE, "%s%04d|Emitting %d bytes for %d x86 bytes", (box86_dynarec_dump>1)?"\e[01;36m":"", GetTID(), helper.arm_size, helper.isize); 
         printFunctionAddr(helper.start, " => ");
         dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");
     }
+    size_t oldarmsize = helper.arm_size;
     helper.arm_size = 0;
     arm_pass3(&helper, addr);
-    if(sz!=helper.arm_size) {
-        printf_log(LOG_NONE, "BOX86: Warning, size difference in block between pass2 (%d) & pass3 (%d)!\n", sz, helper.arm_size);
+    // keep size of instructions for signal handling
+    block->instsize = instsize;
+    // ok, free the helper now
+    customFree(helper.insts);
+    helper.insts = NULL;
+    helper.instsize = NULL;
+    customFree(helper.predecessor);
+    helper.predecessor = NULL;
+    block->size = sz;
+    block->isize = helper.size;
+    block->block = p;
+    block->jmpnext = next+sizeof(void*);
+    block->always_test = helper.always_test;
+    block->dirty = block->always_test;
+    *(dynablock_t**)next = block;
+    *(void**)(next+5*sizeof(void*)) = arm_next;
+    CreateJmpNext(block->jmpnext, next+5*sizeof(void*));
+    //block->x86_addr = (void*)start;
+    block->x86_size = end-start;
+    // all done...
+    __clear_cache(actual_p, actual_p+sz);   // need to clear the cache before execution...
+    block->hash = X31_hash_code(block->x86_addr, block->x86_size);
+    // Check if something changed, to abbort if it as
+    if((block->hash != hash)) {
+        dynarec_log(LOG_DEBUG, "Warning, a block changed while being processed hash(%p:%ld)=%x/%x\n", block->x86_addr, block->x86_size, block->hash, hash);
+        AddHotPage(addr);
+        CancelBlock(0);
+        return NULL;
+    }
+    if((oldarmsize!=helper.arm_size)) {
+        printf_log(LOG_NONE, "BOx86: Warning, size difference in block between pass2 (%zu) & pass3 (%zu)!\n", sz, helper.arm_size);
         uint8_t *dump = (uint8_t*)helper.start;
         printf_log(LOG_NONE, "Dump of %d x86 opcodes:\n", helper.size);
         for(int i=0; i<helper.size; ++i) {
@@ -507,69 +569,16 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
             printf_log(LOG_NONE, "\t%d -> %d\n", helper.insts[i].size2, helper.insts[i].size);
         }
         printf_log(LOG_NONE, " ------------\n");
-    }
-    // all done...
-    __clear_cache(p, p+sz);   // need to clear the cache before execution...
-    // keep size of instructions for signal handling
-    {
-        size_t cap = 1;
-        for(int i=0; i<helper.size; ++i)
-            cap += 1 + ((helper.insts[i].x86.size>helper.insts[i].size)?helper.insts[i].x86.size:helper.insts[i].size)/15;
-        size_t size = 0;
-        block->instsize = (instsize_t*)box_calloc(cap, sizeof(instsize_t));
-        for(int i=0; i<helper.size; ++i)
-            block->instsize = addInst(block->instsize, &size, &cap, helper.insts[i].x86.size, helper.insts[i].size/4);
-        block->instsize = addInst(block->instsize, &size, &cap, 0, 0);    // add a "end of block" mark, just in case
-    }
-    // ok, free the helper now
-    box_free(helper.insts);
-    helper.insts = NULL;
-    box_free(helper.next);
-    helper.next = NULL;
-    block->size = sz;
-    block->isize = helper.size;
-    block->block = p;
-    block->need_test = 0;
-    //block->x86_addr = (void*)start;
-    block->x86_size = end-start;
-    if(box86_dynarec_largest<block->x86_size)
-        box86_dynarec_largest = block->x86_size;
-    block->hash = X31_hash_code(block->x86_addr, block->x86_size);
-    // Check if something changed, to abbort if it as
-    if(block->hash != hash) {
-        dynarec_log(LOG_INFO, "Warning, a block changed while beeing processed hash(%p:%d)=%x/%x\n", block->x86_addr, block->x86_size, block->hash, hash);
-        CancelBlock();
+        CancelBlock(0);
         return NULL;
     }
-    // fill sons if any
-    dynablock_t** sons = NULL;
-    int sons_size = 0;
-    if(helper.sons_size) {
-        sons = (dynablock_t**)box_calloc(helper.sons_size, sizeof(dynablock_t*));
-        for (int i=0; i<helper.sons_size; ++i) {
-            int created = 1;
-            dynablock_t *son = AddNewDynablock(block->parent, helper.sons_x86[i], &created);
-            if(created) {    // avoid breaking a working block!
-                son->block = helper.sons_arm[i];
-                son->x86_addr = (void*)helper.sons_x86[i];
-                son->x86_size = end-helper.sons_x86[i];
-                if(!son->x86_size) {printf_log(LOG_NONE, "Warning, son with null x86 size! (@%p / ARM=%p)", son->x86_addr, son->block);}
-                son->father = block;
-                son->done = 1;
-                sons[sons_size++] = son;
-                if(!son->parent)
-                    son->parent = block->parent;
-            }
-        }
-        if(sons_size) {
-            block->sons = sons;
-            block->sons_size = sons_size;
-        } else
-            box_free(sons);
+    if(!isprotectedDB(addr, end-addr)) {
+        dynarec_log(LOG_DEBUG, "Warning, block unprotected while being processed %p:%ld, marking as need_test\n", block->x86_addr, block->x86_size);
+        AddHotPage(addr);
+        block->dirty = 1;
+        //protectDB(addr, end-addr);
     }
-    box_free(helper.predecessor);
-    helper.predecessor = NULL;
     current_helper = NULL;
-    block->done = 1;
+    //block->done = 1;
     return (void*)block;
 }

src/dynarec/dynarec_arm_00.c

@@ -581,16 +581,12 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0x57:
             INST_NAME("PUSH reg");
             gd = xEAX+(opcode&0x07);
-            #ifdef RPI2
             if(gd==xESP) {
                 MOV_REG(x1, gd);
                 PUSH1(x1);
             } else {
                 PUSH1(gd);    
             }
-            #else
-            PUSH1(gd);
-            #endif
             break;
         case 0x58:
         case 0x59:
@@ -665,7 +661,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             i32 = F32S;
             if(PK(0)==0xC3) {
                 MESSAGE(LOG_DUMP, "PUSH then RET, using indirect\n");
-                MOV32(x3, ip+1);
+                MOV32(x3, addr-4);
                 LDR_IMM9(x1, x3, 0);
                 PUSH1(x1);
             } else {
@@ -1015,7 +1011,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 BFI(eb1, x14, eb2*8, 8);
             } else {
                 GETGB(x14);
-                DMB_ISH();
+                SMDMB();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, LOCK_LOCK);
                 MARKLOCK;
                 // do the swap with exclusive locking
@@ -1025,7 +1021,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x3, 0);
                 B_MARKLOCK(cNE);
                 BFI(gb1, x1, gb2*8, 8);
-                DMB_ISH();
+                SMDMB();
             }
             break;
         case 0x87:
@@ -1041,7 +1037,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 }
             } else {
                 GETGD;
-                DMB_ISH();
+                SMDMB();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 1, LOCK_LOCK);
                 if(!fixedaddress) {
                     TSTS_IMM8(ed, 3);
@@ -1065,7 +1061,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 if(!fixedaddress) {
                     MARK2;
                 }
-                DMB_ISH();
+                SMDMB();
                 MOV_REG(gd, x1);
             }
             break;
@@ -1089,7 +1085,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, &lock);
                 STRB_IMM9(gd, ed, fixedaddress);
-                if(lock) {DMB_ISH();}
+                SMWRITELOCK(lock);
             }
             break;
         case 0x89:
@@ -1101,7 +1097,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {                    // mem <= reg
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, &lock);
                 STR_IMM9(gd, ed, fixedaddress);
-                if(lock) {DMB_ISH();}
+                SMWRITELOCK(lock);
             }
             break;
         case 0x8A:
@@ -1123,10 +1119,10 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 } else {
                     if(box86_dynarec_strongmem && 
                      (dyn->insts[ninst].x86.barrier || !ninst || (box86_dynarec_strongmem>1) || (ninst && dyn->insts[ninst-1].x86.barrier))) {
-                        DMB_ISH();
+                        SMDMB();
                     }
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, &lock);
-                    if(lock) {DMB_ISH();}
+                    SMREADLOCK(lock);
                     LDRB_IMM9(x14, wback, fixedaddress);
                     ed = x14;
                 }
@@ -1140,10 +1136,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 MOV_REG(gd, xEAX+(nextop&7));
             } else {                    // mem <= reg
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, &lock);
-                if(lock || (box86_dynarec_strongmem && 
-                    (dyn->insts[ninst].x86.barrier || !ninst || (box86_dynarec_strongmem>1) || (ninst && dyn->insts[ninst-1].x86.barrier)))) {
-                    DMB_ISH();
-                }
+                SMREADLOCK(lock);
                 LDR_IMM9(gd, ed, fixedaddress);
             }
             break;
@@ -1157,6 +1150,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 LDRH_REG(x3, xEmu, x3);
                 STRH_IMM8(x3, ed, fixedaddress);
+                SMWRITE2();
             }
             break;
         case 0x8D:
@@ -1284,12 +1278,14 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             u32 = F32;
             MOV32(x2, u32);
             STRB_IMM9(xEAX, x2, 0);
+            SMWRITE();
             break;
         case 0xA3:
             INST_NAME("MOV Od, EAX");
             u32 = F32;
             MOV32(x2, u32);
             STR_IMM9(xEAX, x2, 0);
+            SMWRITE();
             break;
         case 0xA4:
             INST_NAME("MOVSB");
@@ -1522,7 +1518,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     u8 = F8;
                     MOVW(x2, u8);
                     CALL_(rcl32, ed, (1<<x14));
-                    WBACK;  // no SBACK here because ret value is ed directly
+                    WBACK;
                     break;
                 case 3:
                     INST_NAME("RCR Ed, Ib");
@@ -1553,13 +1549,13 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             BFC_COND(cCC, xFlags, F_CF, 1);
 
                         }
-                        WBACK;  // no SBACK here because ed has been used all along and not x1
+                        WBACK;
                     } else if(u8>1) {
                         MESSAGE(LOG_DUMP, "Need Optimization RCR\n");
                         SETFLAGS(X_OF|X_CF, SF_SET);
                         MOVW(x2, u8);
                         CALL_(rcr32, ed, (1<<x14));
-                        WBACK;  // no SBACK here because ret value is ed directly
+                        WBACK;
                     }
                     break;
                 case 4:
@@ -1633,7 +1629,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 u8 = F8;
                 MOVW(x3, u8);
                 STRB_IMM9(x3, ed, fixedaddress);
-                if(lock) {DMB_ISH();}
+                SMWRITELOCK(lock);
             }
             break;
         case 0xC7:
@@ -1648,7 +1644,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 i32 = F32S;
                 MOV32(x3, i32);
                 STR_IMM9(x3, ed, fixedaddress);
-                if(lock) {DMB_ISH();}
+                SMWRITELOCK(lock);
             }
             break;
 
@@ -1667,6 +1663,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 if((PK(0)==0) && (PK(1)==0) && (PK(2)==0) && (PK(3)==0))
                 {
                     addr+=4;
+                    SKIPTEST(x14);
                     MESSAGE(LOG_DEBUG, "Exit x86 Emu\n");
                     MOV32(x14, ip+1+2);
                     STM(xEmu, (1<<xEAX)|(1<<xEBX)|(1<<xECX)|(1<<xEDX)|(1<<xESI)|(1<<xEDI)|(1<<xESP)|(1<<xEBP)|(1<<xEIP)|(1<<xFlags));
@@ -1676,6 +1673,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     *need_epilog = 1;
                 } else {
                     MESSAGE(LOG_DUMP, "Native Call to %s\n", GetNativeName(GetNativeFnc(ip)));
+                    SKIPTEST(x14);
                     x87_forget(dyn, ninst, x3, x14, 0);
                     if((box86_log<2) && !cycle_log) {   // call the wrapper directly
                         uintptr_t ncall[2]; // to avoid BUSERROR!!!
@@ -1719,9 +1717,11 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xCD:
             SETFLAGS(X_ALL, SF_SET);    // Hack, set all flags (to an unknown state...)
-            if(PK(0)==0x80) {
+            SKIPTEST(x14);
+            SMEND();
+            u8 = F8;
+            if(u8==0x80) {
                 INST_NAME("Syscall");
-                u8 = F8;
                 BARRIER(BARRIER_FLOAT);
                 MOV32(xEIP, ip+2);
                 STM(xEmu, (1<<xEAX)|(1<<xEBX)|(1<<xECX)|(1<<xEDX)|(1<<xESI)|(1<<xEDI)|(1<<xESP)|(1<<xEBP)|(1<<xEIP)|(1<<xFlags));
@@ -1737,7 +1737,11 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 jump_to_epilog(dyn, 0, xEIP, ninst);
             } else {
                 INST_NAME("INT Ib");
-                DEFAULT;
+                if(box86_wine && u8==0x2D) {
+                    MESSAGE(LOG_DEBUG, "Hack for wine/int 2d\n");
+                } else {
+                    DEFAULT;
+                }
             }
             break;
 
@@ -1884,7 +1888,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     MOVW(x2, 1);
                     GETEDW(x14, x1);
                     CALL_(rcl32, ed, (1<<x14));
-                    WBACK;  // no SBACK here because ret value is ed directly
+                    WBACK;
                     break;
                 case 3:
                     INST_NAME("RCR Ed, 1");
@@ -1994,7 +1998,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     AND_IMM8(x2, xECX, 0x1f);
                     GETEDW(x14, x1);
                     CALL_(rcl32, ed, (1<<x14));
-                    WBACK;  // no SBACK here because ret value is ed directly
+                    WBACK;
                     break;
                 case 3:
                     INST_NAME("RCR Ed, CL");
@@ -2004,7 +2008,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     AND_IMM8(x2, xECX, 0x1f);
                     GETEDW(x14, x1);
                     CALL_(rcr32, ed, (1<<x14));
-                    WBACK;  // no SBACK here because ret value is ed directly
+                    WBACK;
                     break;
                 case 4:
                 case 6:
@@ -2065,6 +2069,7 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0xD7:
             INST_NAME("XLAT");
             UBFX(x1, xEAX, 0, 8);    // x1 = AL
+            SMREAD();
             LDRB_REG_LSL_IMM5(x1, xEBX, x1, 0); //x1 = byte ptr[EBX+AL]
             BFI(xEAX, x1, 0, 8);
             break;
@@ -2160,8 +2165,8 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 tmp = dyn->insts[ninst].pass2choice = 1;
             else if ((getProtection(u32)&PROT_READ) && (PKa(u32+0)==0x8B) && (((PKa(u32+1))&0xC7)==0x04) && (PKa(u32+2)==0x24) && (PKa(u32+3)==0xC3))
                 tmp = dyn->insts[ninst].pass2choice = 2;
-            /*else if(isNativeCall(dyn, u32, &dyn->insts[ninst].natcall, &dyn->insts[ninst].retn))
-                tmp = dyn->insts[ninst].pass2choice = 3;*/
+            else if(isNativeCall(dyn, u32, &dyn->insts[ninst].natcall, &dyn->insts[ninst].retn))
+                tmp = dyn->insts[ninst].pass2choice = 3;
             else 
                 tmp = dyn->insts[ninst].pass2choice = 0;
             #else
@@ -2171,18 +2176,19 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 case 1:
                     //SETFLAGS(X_ALL, SF_SET);    // Hack to set flags to "dont'care" state
                     MESSAGE(LOG_DUMP, "Hack for Call 0\n");
+                    SKIPTEST(x14);
                     MOV32(xEIP, addr);
                     PUSH1(xEIP);
                     break;
                 case 2:
                     //SETFLAGS(X_ALL, SF_SET);    // Hack to set flags to "dont'care" state
                     MESSAGE(LOG_DUMP, "Hack for Call x86.get_pc_thunk.reg\n");
+                    SKIPTEST(x14);
                     u8 = PK(i32+1);
                     gd = xEAX+((u8&0x38)>>3);
                     MOV32(gd, addr);
                     break;
-                // disabling this to avoid fetching data outside current block (in case this part changed, this block will not been marck as dirty)
-                /*case 3:
+                case 3:
                     SETFLAGS(X_ALL, SF_SET);    // Hack to set flags to "dont'care" state
                     BARRIER(BARRIER_FULL);
                     BARRIER_NEXT(BARRIER_FULL);
@@ -2221,16 +2227,22 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     B_NEXT(cNE);    // not quitting, so lets continue
                     MARK;
                     jump_to_epilog(dyn, 0, xEIP, ninst);
-                    break;*/
+                    break;
                 default:
                     if((box86_dynarec_safeflags>1) || (ninst && dyn->insts[ninst-1].x86.set_flags)) {
                         READFLAGS(X_PEND);  // that's suspicious
                     } else {
-                        //SETFLAGS(X_ALL, SF_SET);    // Hack to set flags to "dont'care" state
+                        SETFLAGS(X_ALL, SF_SET);    // Hack to set flags to "dont'care" state
                     }
                     // regular call
-                    BARRIER(BARRIER_FLOAT);
                     //BARRIER_NEXT(BARRIER_FULL);
+                    if(/*box86_dynarec_callret &&*/ box86_dynarec_bigblock>1) {
+                        BARRIER(BARRIER_FULL);
+                    } else {
+                        BARRIER(BARRIER_FLOAT);
+                        *need_epilog = 0;
+                        *ok = 0;
+                    }
                     *need_epilog = 0;
                     *ok = 0;
                     MOV32(x2, addr);
@@ -2825,10 +2837,13 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 2: // CALL Ed
                     INST_NAME("CALL Ed");
-                    PASS2IF((box86_dynarec_safeflags>1) || (ninst && dyn->insts[ninst-1].x86.set_flags), 1) {
+                    PASS2IF((box86_dynarec_safeflags>1) || 
+                        ((ninst && dyn->insts[ninst-1].x86.set_flags)
+                        || ((ninst>1) && dyn->insts[ninst-2].x86.set_flags)), 1)
+                    {
                         READFLAGS(X_PEND);          // that's suspicious
                     } else {
-                        //SETFLAGS(X_ALL, SF_SET);    //Hack to put flag in "don't care" state
+                        SETFLAGS(X_ALL, SF_SET);    //Hack to put flag in "don't care" state
                     }
                     GETEDH(xEIP);
                     BARRIER(BARRIER_FLOAT);
@@ -2841,7 +2856,8 @@ uintptr_t dynarec00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 4: // JMP Ed
                     INST_NAME("JMP Ed");
-                    BARRIER(BARRIER_FULL);
+                    READFLAGS(X_PEND);
+                    BARRIER(BARRIER_FLOAT);
                     GETEDH(xEIP);
                     jump_to_next(dyn, 0, ed, ninst);
                     *need_epilog = 0;

src/dynarec/dynarec_arm_0f.c

@@ -31,6 +31,7 @@
     if((nextop&0xC0)==0xC0) { \
         a = sse_get_reg(dyn, ninst, x1, nextop&7, w); \
     } else {    \
+        SMREAD(); \
         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL); \
         a = fpu_get_scratch_quad(dyn); \
         VLD1Q_64(a, ed);    \
@@ -42,6 +43,7 @@
     if((nextop&0xC0)==0xC0) { \
         a = mmx_get_reg(dyn, ninst, x1, x2, x3, nextop&7); \
     } else {    \
+        SMREAD(); \
         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL); \
         a = fpu_get_scratch_double(dyn); \
         VLD1_64(a, ed);    \
@@ -49,6 +51,7 @@
 #define PUTEM(a)    \
     if((nextop&0xC0)!=0xC0) { \
         VST1_64(a, ed);    \
+        SMWRITE2(); \
     }
 
 uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
@@ -87,6 +90,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             INST_NAME("UD2");
             SETFLAGS(X_ALL, SF_SET);    // Hack to set flags in "don't care" state
             //CALL(arm_ud, -1, 0);
+            SKIPTEST(x14);
             UDF(0);
             break;
 
@@ -112,6 +116,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
                 //LDRD also have alignment requirements
@@ -140,6 +145,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 VMOVfrV_D(x2, x3, v0+1);
                 STR_IMM9(x2, ed, fixedaddress+8);
                 STR_IMM9(x3, ed, fixedaddress+12);
+                SMWRITE2();
             }
             break;
         case 0x12:
@@ -152,6 +158,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {
                 INST_NAME("MOVLPS Gx,Ex");
                 GETGX(v0, 1);
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress);
                 LDR_IMM9(x3, ed, fixedaddress+4);
@@ -170,12 +177,13 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 if(parity) {
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                     VST1_32(v0, ed);  // better to use VST1 than VSTR_64, to avoid NEON->VFPU transfert I assume
-
+                    SMWRITE2();
                 } else {
                     VMOVfrV_D(x2, x3, v0);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                     STR_IMM9(x2, ed, fixedaddress);
                     STR_IMM9(x3, ed, fixedaddress+4);
+                    SMWRITE2();
                 }
             }
             break;
@@ -212,6 +220,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {
                 INST_NAME("MOVHPS Gx,Ex");
                 GETGX(v0, 1);
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 VLD1_64(v0+1, ed);
             }
@@ -228,12 +237,13 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 if(parity) {
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                     VST1_64(v0+1, ed);  // better to use VST1 than VSTR_64, to avoid NEON->VFPU transfert I assume
-
+                    SMWRITE2();
                 } else {
                     VMOVfrV_D(x2, x3, v0+1);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                     STR_IMM9(x2, ed, fixedaddress);
                     STR_IMM9(x3, ed, fixedaddress+4);
+                    SMWRITE2();
                 }
             }
             break;
@@ -278,6 +288,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 VMOVQ(v0, v1);
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 VLD1Q_32(v0, ed);
             }
@@ -293,6 +304,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 VST1Q_32(v0, ed);
+                SMWRITE2();
             }
             break;
         case 0x2A:
@@ -303,6 +315,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             if((nextop&0xC0)==0xC0) {
                 v1 = mmx_get_reg(dyn, ninst, x1, x2, x3, nextop&7);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_double(dyn);
                 VLD1_32(v1, ed);
@@ -330,6 +343,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             if((nextop&0xC0)==0xC0) {
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_double(dyn);
                 VLD1_32(v1, ed);
@@ -344,6 +358,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             if((nextop&0xC0)==0xC0) {
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_double(dyn);
                 VLD1_32(v1, ed);
@@ -383,6 +398,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 }
                 s0 = d0*2;
             } else {
+                SMREAD();
                 parity = getedparity(dyn, ninst, addr, nextop, 2);
                 s0 = fpu_get_scratch_single(dyn);
                 if(parity) {
@@ -406,6 +422,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0x31:
             INST_NAME("RDTSC");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            SKIPTEST(x14);
             CALL(ReadTSC, xEAX, 0);   // will return the u64 in x1:xEAX
             MOV_REG(xEDX, x1);
             break;
@@ -452,6 +469,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {   // reg <= reg
                         REV(gd, xEAX+(nextop&7));
                     } else {                    // mem <= reg
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, NULL);
                         LDR_IMM9(gd, ed, fixedaddress);
                         REV(gd, gd);
@@ -467,6 +485,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, NULL);
                         REV(x1, gd);
                         STR_IMM9(gd, ed, fixedaddress);
+                        SMWRITE();
                     }
                     break;
 
@@ -508,6 +527,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 ed = xEAX+(nextop&7);   \
                 MOV_REG_COND(YES, gd, ed); \
             } else { \
+                SMREAD(); \
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, NULL);    \
                 LDR_IMM9_COND(YES, gd, ed, fixedaddress); \
             }
@@ -620,6 +640,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 BFI(gd, x3, 3, 1);
             } else {
                 // EX is memory
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4096-16, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
                 UBFX(x3, x2, 31, 1);
@@ -637,6 +658,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0x51:
             INST_NAME("SQRTPS Gx, Ex");
+            SKIPTEST(x1);
             nextop = F8;
             GETEX(q0, 0);
             gd = (nextop&0x38)>>3;
@@ -676,6 +698,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0x52:
             INST_NAME("RSQRTPS Gx, Ex");
+            SKIPTEST(x1);
             nextop = F8;
             GETEX(q0, 0);
             gd = (nextop&0x38)>>3;
@@ -684,6 +707,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0x53:
             INST_NAME("RCPPS Gx, Ex");
+            SKIPTEST(x1);
             nextop = F8;
             GETEX(q0, 0);
             gd = (nextop&0x38)>>3;
@@ -772,7 +796,23 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             nextop = F8;
             GETGX(v0, 1);
             GETEX(v1, 0);
+            // VMIN/VMAX wll put default NaN if any input is NaN
+            // but x86 will copy if either v0[x] or v1[x] is NaN, so lets force a copy if source is NaN
+            if(!box86_dynarec_fastnan) {
+                if(v0==v1) {
+                    q1 = fpu_get_scratch_quad(dyn);
+                    VMOVQ(q1, v1);
+                } else
+                    q1 = v1;
+            }
             VMINQ_F32(v0, v0, v1);
+            if(!box86_dynarec_fastnan) {
+                q0 = fpu_get_scratch_quad(dyn);
+                VCEQQ_F32(q0, v0, v0);   // 0 is NaN, -1 is not NaN
+                VANDQ(v0, v0, q0);
+                VBICQ(q0, q1, q0);
+                VORRQ(v0, v0, q0);
+            }
             break;
         case 0x5E:
             INST_NAME("DIVPS Gx, Ex");
@@ -796,7 +836,23 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             nextop = F8;
             GETGX(v0, 1);
             GETEX(v1, 0);
+            // VMIN/VMAX wll put default NaN if any input is NaN
+            // but x86 will copy if either v0[x] or v1[x] is NaN, so lets force a copy if source is NaN
+            if(!box86_dynarec_fastnan) {
+                if(v0==v1) {
+                    q1 = fpu_get_scratch_quad(dyn);
+                    VMOVQ(q1, v1);
+                } else
+                    q1 = v1;
+            }
             VMAXQ_F32(v0, v0, v1);
+            if(!box86_dynarec_fastnan) {
+                q0 = fpu_get_scratch_quad(dyn);
+                VCEQQ_F32(q0, v0, v0);   // 0 is NaN, -1 is not NaN
+                VANDQ(v0, v0, q0);
+                VBICQ(q0, q1, q0);
+                VORRQ(v0, v0, q0);
+            }
             break;
         case 0x60:
             INST_NAME("PUNPCKLBW Gm,Em");
@@ -1005,6 +1061,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {
                         d0 = mmx_get_reg(dyn, ninst, x1, x2, x3, nextop&7);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         d0 = fpu_get_scratch_quad(dyn);
                         VLD1_16(d0, ed);
@@ -1018,6 +1075,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1_16(d0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1026,6 +1084,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {
                         d0 = mmx_get_reg(dyn, ninst, x1, x2, x3, nextop&7);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         d0 = fpu_get_scratch_quad(dyn);
                         VLD1_16(d0, ed);
@@ -1036,6 +1095,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     }
                     if((nextop&0xC0)!=0xC0) {
                         VST1_16(d0, ed);
+                        SMWRITE2();
                     }
                     break;
                 case 6:
@@ -1043,6 +1103,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {
                         d0 = mmx_get_reg(dyn, ninst, x1, x2, x3, nextop&7);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         d0 = fpu_get_scratch_quad(dyn);
                         VLD1_16(d0, ed);
@@ -1056,6 +1117,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1_16(d0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1065,6 +1127,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0x72:
             nextop = F8;
+            SMREAD();
             GETEM(v0);
             switch((nextop>>3)&7) {
                 case 2:
@@ -1106,6 +1169,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0x73:
             nextop = F8;
             GETEM(v0);
+            SMREAD();
             switch((nextop>>3)&7) {
                 case 2:
                     INST_NAME("PSRLQ Em, Ib");
@@ -1171,6 +1235,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 VMOVfrDx_32(x2, v0, 0); // there can be some bus error is storing the VFPU reg directly
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0, NULL);
                 STR_IMM9(x2, ed, fixedaddress);
+                SMWRITE2();
             }
             break;
         case 0x7F:
@@ -1185,11 +1250,13 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 if(parity) {
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                     VSTR_64(v0, ed, fixedaddress);
+                    SMWRITE2();
                 } else {
                     VMOVfrV_D(x2, x3, v0);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                     STR_IMM9(x2, ed, fixedaddress);
                     STR_IMM9(x3, ed, fixedaddress+4);
+                    SMWRITE2();
                 }
             }
             break;
@@ -1307,6 +1374,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 , cNE, cEQ, X_SF|X_OF|X_ZF)
             break;
         #undef GO
+
         #define GO(GETFLAGS, NO, YES, F)    \
             READFLAGS(F);                   \
             GETFLAGS;   \
@@ -1321,8 +1389,8 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {                \
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, NULL);\
                 STRB_IMM9(x3, ed, fixedaddress);   \
+                SMWRITE();          \
             }
-
         case 0x90:
             INST_NAME("SETO Eb");
             GO( TSTS_IMM8_ROR(xFlags, 0b10, 0x0b)
@@ -1443,6 +1511,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, 4095, 0, 0, NULL);
                 MOV_REG_ASR_IMM5(x1, gd, 5);    // r1 = (gd>>5);
                 ADD_REG_LSL_IMM5(x1, ed, x1, 2); //(&ed)+=r1*4;
@@ -1499,6 +1568,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 ed = xEAX+(nextop&7);
                 wback = 0;
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL);
                 MOV_REG_ASR_IMM5(x1, gd, 5);    // r1 = (gd>>5);
                 ADD_REG_LSL_IMM5(x3, wback, x1, 2); //(&ed)+=r1*4;
@@ -1517,6 +1587,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             ORR_REG_LSL_REG(ed, ed, x14, x2);
             if(wback) {
                 STR_IMM9(ed, wback, fixedaddress);
+                SMWRITE();
             }
             break;
         case 0xAC:
@@ -1544,15 +1615,15 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             // xFENCE -> DMB is approximative, but should be good enough
             if((nextop&0xF8)==0xE8) {
                 INST_NAME("LFENCE");
-                DMB_ISH();
+                SMDMB();
             } else
             if((nextop&0xF8)==0xF0) {
                 INST_NAME("MFENCE");
-                DMB_ISH();
+                SMDMB();
             } else
             if((nextop&0xF8)==0xF8) {
                 INST_NAME("SFENCE");
-                DMB_ISH();
+                SMDMB();
             } else {
                 switch((nextop>>3)&7) {
                     case 0:
@@ -1584,6 +1655,12 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         INST_NAME("LDMXCSR Md");
                         GETED;
                         STR_IMM9(ed, xEmu, offsetof(x86emu_t, mxcsr));
+                        if(box86_sse_flushto0) {
+                            VMRS(x14);                      // get fpscr
+                            MOV_REG_LSR_IMM5(x3, ed, 15);   // get FZ bit
+                            BFI(x14, x3, 24, 1);            // inject FZ bit
+                            VMSR(x14);                      // put new fpscr
+                        }
                         break;
                     case 3:
                         INST_NAME("STMXCSR Md");
@@ -1674,6 +1751,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 ed = xEAX+(nextop&7);
                 wback = 0;
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL);
                 MOV_REG_ASR_IMM5(x1, gd, 5);    // r1 = (gd>>5);
                 ADD_REG_LSL_IMM5(x3, wback, x1, 2); //(&ed)+=r1*4;
@@ -1692,6 +1770,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             XOR_REG_LSL_REG(ed, ed, x14, x2);
             if(wback) {
                 STR_IMM9(ed, wback, fixedaddress);
+                SMWRITE();
             }
             MARK3;
             break;
@@ -1706,6 +1785,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 eb2 = (ed&4)>>2;    // L or H
                 UXTB(gd, eb1, eb2);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, NULL);
                 LDRB_IMM9(gd, ed, fixedaddress);
             }
@@ -1718,6 +1798,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 ed = xEAX+(nextop&7);
                 UXTH(gd, ed, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 LDRH_IMM8(gd, ed, fixedaddress);
             }
@@ -1734,6 +1815,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL);
                         LDR_IMM9(x1, wback, fixedaddress);
                         ed = x1;
@@ -1754,6 +1836,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         ed = xEAX+(nextop&7);
                         wback = 0;
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL);
                         LDR_IMM9(x1, wback, fixedaddress);
                         ed = x1;
@@ -1774,6 +1857,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     XOR_REG_LSL_IMM5(ed, ed, x14, u8);
                     if(wback) {
                         STR_IMM9(ed, wback, fixedaddress);
+                        SMWRITE();
                     }
                     break;
                 case 6:
@@ -1784,6 +1868,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         ed = xEAX+(nextop&7);
                         wback = 0;
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL);
                         LDR_IMM9(x1, wback, fixedaddress);
                         ed = x1;
@@ -1804,6 +1889,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     XOR_REG_LSL_IMM5(ed, ed, x14, u8);
                     if(wback) {
                         STR_IMM9(ed, wback, fixedaddress);
+                        SMWRITE();
                     }
                     break;
                 case 7:
@@ -1814,6 +1900,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         ed = xEAX+(nextop&7);
                         wback = 0;
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL);
                         LDR_IMM9(x1, wback, fixedaddress);
                         ed = x1;
@@ -1835,6 +1922,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     XOR_REG_LSL_IMM5(ed, ed, x14, u8);
                     if(wback) {
                         STR_IMM9(ed, wback, fixedaddress);
+                        SMWRITE();
                     }
                     MARK3;
                     break;
@@ -1852,6 +1940,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 ed = xEAX+(nextop&7);
                 wback = 0;
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL);
                 MOV_REG_ASR_IMM5(x1, gd, 5);    // r1 = (gd>>5);
                 ADD_REG_LSL_IMM5(x3, wback, x1, 2); //(&ed)+=r1*4;
@@ -1868,6 +1957,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             XOR_REG_LSL_REG(ed, ed, x14, x2);
             if(wback) {
                 STR_IMM9(ed, wback, fixedaddress);
+                SMWRITE();
             }
             break;
         case 0xBC:
@@ -1924,6 +2014,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 eb2 = (ed&4)>>2;    // L or H
                 SXTB(gd, eb1, eb2);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 LDRSB_IMM8(gd, ed, fixedaddress);
             }
@@ -1936,6 +2027,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 ed = xEAX+(nextop&7);
                 SXTH(gd, ed, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 LDRSH_IMM8(gd, ed, fixedaddress);
             }
@@ -1956,12 +2048,11 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             nextop = F8;
             GETGD;
             GETED;
+            MOV_REG(x1, gd);
             if(gd!=ed) {
-                XOR_REG_LSL_IMM5(gd, gd, ed, 0);    // swap gd, ed
-                XOR_REG_LSL_IMM5(ed, gd, ed, 0);
-                XOR_REG_LSL_IMM5(gd, gd, ed, 0);
+                MOV_REG(gd, ed);
             }
-            emit_add32(dyn, ninst, ed, gd, x3, x14);
+            emit_add32(dyn, ninst, ed, x1, x3, x14);
             WBACK;
             break;
         case 0xC2:
@@ -2016,6 +2107,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 ed = xEAX+(nextop&7);
                 VMOVtoDx_16(d0, u8, ed);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0, NULL);
                 u8 = (F8)&3;
                 VLD1LANE_16(d0, wback, u8);
@@ -2030,6 +2122,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 u8 = (F8)&3;
                 VMOVfrDx_U16(gd, d0, u8);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255-8, 0, 0, NULL);
                 u8 = (F8)&3;
                 LDRH_IMM8(gd, wback, fixedaddress+u8*2);
@@ -2043,6 +2136,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             GETGX(v0, 1);
             v2 = fpu_get_scratch_double(dyn);
             if((nextop&0xC0)!=0xC0) {
+                SMREAD();
                 // from memory
                 addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 0, 0, 0, NULL);
                 u8 = F8;
@@ -2194,6 +2288,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 case 1:
                     INST_NAME("CMPXCHG8B Gq, Eq");
                     SETFLAGS(X_ZF, SF_SUBSET);
+                    SMREAD();
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095-4, 0, 0, NULL);
                     LDR_IMM9(x1, wback, fixedaddress+0);
                     LDR_IMM9(x2, wback, fixedaddress+4);
@@ -2211,6 +2306,7 @@ uintptr_t dynarec0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     MOVW(x1, 0);
                     MARK3;
                     BFI(xFlags, x1, F_ZF, 1);
+                    SMWRITE();
                     break;
                 default:
                     DEFAULT;

src/dynarec/dynarec_arm_64.c

@@ -103,6 +103,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     } else {
                         addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                         STRH_REG(gd, x14, ed);
+                        SMWRITE2();
                     }
                     break;
 
@@ -114,6 +115,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {   // reg <= reg
                         MOV_REG(gd, xEAX+(nextop&7));
                     } else {                    // mem <= reg
+                        SMREAD();
                         addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                         LDR_REG_LSL_IMM5(gd, ed, x14, 0);
                     }
@@ -129,6 +131,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         addr = geted16(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
                         POP1(x2);
                         STR_REG_LSL_IMM5(x2, x1, x14, 0);
+                        SMWRITE2();
                     }
                     break;
 
@@ -140,6 +143,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         MOV32(x2, u32);
                         ADD_REG_LSL_IMM5(x1, x1, x2, 0);
                     }
+                    SMREAD();
                     LDR_IMM9(xEAX, x1, 0);
                     break;
 
@@ -152,6 +156,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         ADD_REG_LSL_IMM5(x1, x1, x2, 0);
                     }
                     STR_IMM9(xEAX, x1, 0);
+                    SMWRITE2();
                     break;
 
                 case 0xFF:
@@ -163,6 +168,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             if((nextop&0xC0)==0xC0) {   // reg
                                 DEFAULT;
                             } else {                    // mem <= i32
+                                SMREAD();
                                 addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                                 LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                                 PUSH1(x3);
@@ -187,6 +193,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {                    // mem <= reg
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 STR_REG_LSL_IMM5(gd, ed, x14, 0);
+                SMWRITE2();
             }
             break;
 
@@ -198,6 +205,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             if((nextop&0xC0)==0xC0) {   // reg <= reg
                 MOV_REG(gd, xEAX+(nextop&7));
             } else {                    // mem <= reg
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 LDR_REG_LSL_IMM5(gd, ed, x14, 0);
             }
@@ -213,6 +221,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 POP1(x2);
                 STR_REG_LSL_IMM5(x2, ed, x14, 0);
+                SMWRITE2();
             }
             break;
 
@@ -224,6 +233,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 LDRB_IMM9(xEAX, x1, i32);
             } else {
                 MOV32(x2, i32);
+                SMREAD();
                 LDRB_REG_LSL_IMM5(xEAX, x1, x2, 0);
             }
             break;
@@ -235,6 +245,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 LDR_IMM9(xEAX, x1, i32);
             } else {
                 MOV32(x2, i32);
+                SMREAD();
                 LDR_REG_LSL_IMM5(xEAX, x1, x2, 0);
             }
             break;
@@ -247,6 +258,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {
                 MOV32(x2, i32);
                 STRB_REG_LSL_IMM5(xEAX, x1, x2, 0);
+                SMWRITE2();
             }
             break;
         case 0xA3:
@@ -255,9 +267,11 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             i32 = F32S;
             if(i32>-4096 && i32<4096) {
                 STR_IMM9(xEAX, x1, i32);
+                SMWRITE2();
             } else {
                 MOV32(x2, i32);
                 STR_REG_LSL_IMM5(xEAX, x2, x1, 0);
+                SMWRITE2();
             }
             break;
 
@@ -283,6 +297,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 i32 = F32S;
                 MOV32(x3, i32);
                 STR_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
             }
             break;
             
@@ -298,7 +313,10 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             switch((nextop>>3)&7) {
                 case 2: // CALL Ed
                     INST_NAME("CALL FS:Ed");
-                    PASS2IF(ninst && dyn->insts[ninst-1].x86.set_flags, 1) {
+                    PASS2IF((box86_dynarec_safeflags>1) ||
+                        ((ninst && dyn->insts[ninst-1].x86.set_flags)
+                        || ((ninst>1) && dyn->insts[ninst-2].x86.set_flags)), 1)
+                    {
                         READFLAGS(X_PEND);          // that's suspicious
                     } else {
                         SETFLAGS(X_ALL, SF_SET);    //Hack to put flag in "don't care" state
@@ -321,6 +339,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {   // reg
                         DEFAULT;
                     } else {                    // mem <= i32
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                         LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                         PUSH1(x3);

src/dynarec/dynarec_arm_65.c

@@ -50,7 +50,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             GETGD;
             GETEDO2(x14);
             emit_add32(dyn, ninst, ed, gd, x3, x14);
-            WBACK;
+            WBACK2;
             break;
 
         case 0x03:
@@ -97,7 +97,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             GETGD;
             GETEDO2(x14);
             emit_and32(dyn, ninst, ed, gd, x3, x14);
-            WBACK;
+            WBACK2;
             break;
 
         case 0x23:
@@ -118,7 +118,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             GETGD;
             GETEDO2(x14);
             emit_sub32(dyn, ninst, ed, gd, x3, x14);
-            WBACK;
+            WBACK2;
             break;
 
         case 0x2B:
@@ -139,7 +139,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             GETGD;
             GETEDO2(x14);
             emit_xor32(dyn, ninst, gd, ed, x3, x14);
-            WBACK;
+            WBACK2;
             break;
 
         case 0x33:
@@ -202,7 +202,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEBO(x1);
                     u8 = F8;
                     emit_add8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                     break;
                 case 1: //OR
                     INST_NAME("OR GS:Eb, Ib");
@@ -211,7 +211,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEBO(x1);
                     u8 = F8;
                     emit_or8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                     break;
                 case 2: //ADC
                     INST_NAME("ADC GS:Eb, Ib");
@@ -221,7 +221,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEBO(x1);
                     u8 = F8;
                     emit_adc8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                     break;
                 case 3: //SBB
                     INST_NAME("SBB GS:Eb, Ib");
@@ -231,7 +231,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEBO(x1);
                     u8 = F8;
                     emit_sbb8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                     break;
                 case 4: //AND
                     INST_NAME("AND GS:Eb, Ib");
@@ -240,7 +240,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEBO(x1);
                     u8 = F8;
                     emit_and8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                     break;
                 case 5: //SUB
                     INST_NAME("SUB GS:Eb, Ib");
@@ -249,7 +249,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEBO(x1);
                     u8 = F8;
                     emit_sub8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                     break;
                 case 6: //XOR
                     INST_NAME("XOR GS:Eb, Ib");
@@ -258,7 +258,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEBO(x1);
                     u8 = F8;
                     emit_xor8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                     break;
                 case 7: //CMP
                     INST_NAME("CMP GS:Eb, Ib");
@@ -290,7 +290,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
                     emit_add32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 1: //OR
                     if(opcode==0x81) {INST_NAME("OR Ed, Id");} else {INST_NAME("OR Ed, Ib");}
@@ -299,7 +299,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
                     emit_or32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 2: //ADC
                     if(opcode==0x81) {INST_NAME("ADC Ed, Id");} else {INST_NAME("ADC Ed, Ib");}
@@ -309,7 +309,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
                     emit_adc32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 3: //SBB
                     if(opcode==0x81) {INST_NAME("SBB Ed, Id");} else {INST_NAME("SBB Ed, Ib");}
@@ -319,7 +319,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
                     emit_sbb32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 4: //AND
                     if(opcode==0x81) {INST_NAME("AND Ed, Id");} else {INST_NAME("AND Ed, Ib");}
@@ -328,7 +328,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
                     emit_and32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 5: //SUB
                     if(opcode==0x81) {INST_NAME("SUB Ed, Id");} else {INST_NAME("SUB Ed, Ib");}
@@ -337,7 +337,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
                     emit_sub32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 6: //XOR
                     if(opcode==0x81) {INST_NAME("XOR Ed, Id");} else {INST_NAME("XOR Ed, Ib");}
@@ -346,7 +346,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
                     emit_xor32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 7: //CMP
                     if(opcode==0x81) {INST_NAME("CMP Ed, Id");} else {INST_NAME("CMP Ed, Ib");}
@@ -374,6 +374,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {                    // mem <= reg
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 STR_REG_LSL_IMM5(gd, ed, x14, 0);
+                SMWRITE2();
             }
             break;
 
@@ -385,6 +386,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             if((nextop&0xC0)==0xC0) {   // reg <= reg
                 MOV_REG(gd, xEAX+(nextop&7));
             } else {                    // mem <= reg
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 LDR_REG_LSL_IMM5(gd, ed, x14, 0);
             }
@@ -400,6 +402,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 POP1(x2);
                 STR_REG_LSL_IMM5(x2, ed, x14, 0);
+                SMWRITE2();
             }
             break;
 
@@ -407,6 +410,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             INST_NAME("MOV EAX, GS:Id");
             grab_tlsdata(dyn, addr, ninst, x1);
             i32 = F32S;
+            SMREAD();
             if(i32>-4096 && i32<4096) {
                 LDR_IMM9(xEAX, x1, i32);
             } else {
@@ -423,6 +427,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             MOV32(x2, u32);
             ADD_REG_LSL_IMM5(x2, x1, x2, 0);
             STRB_IMM9(xEAX, x2, 0);
+            SMWRITE2();
             break;
 
         case 0xA3:
@@ -432,6 +437,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             MOV32(x2, u32);
             ADD_REG_LSL_IMM5(x2, x1, x2, 0);
             STR_IMM9(xEAX, x2, 0);
+            SMWRITE2();
             break;
 
         case 0xC6:
@@ -450,6 +456,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 u8 = F8;
                 MOVW(x3, u8);
                 STRB_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
             }
             break;
         case 0xC7:
@@ -465,6 +472,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 i32 = F32S;
                 MOV32(x3, i32);
                 STR_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
             }
             break;
 
@@ -477,14 +485,14 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETEDO2(x14);
                     emit_rol32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 1:
                     INST_NAME("ROR Ed, 1");
                     SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETEDO2(x14);
                     emit_ror32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 2:
                     INST_NAME("RCL Ed, 1");
@@ -493,10 +501,9 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     SETFLAGS(X_OF|X_CF, SF_SET);
                     MOVW(x2, 1);
                     GETEDO2(x14);
-                    wb = ed;
                     if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
-                    CALL_(rcl32, ed, (1<<x14));
-                    SBACK(wb);
+                    CALL_(rcl32, ed, (1<<x2));
+                    WBACK2;
                     break;
                 case 3:
                     INST_NAME("RCR Ed, 1");
@@ -504,10 +511,9 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     READFLAGS(X_CF);
                     SETFLAGS(X_OF|X_CF, SF_SET);
                     MOVW(x2, 1);
-                    wb = ed;
                     if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
-                    CALL_(rcr32, ed, (1<<x14));
-                    SBACK(wb);
+                    CALL_(rcr32, ed, (1<<x2));
+                    WBACK2;
                     break;
                 case 4:
                 case 6:
@@ -515,21 +521,21 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                     GETEDO2(x14);
                     emit_shl32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 5:
                     INST_NAME("SHR Ed, 1");
                     SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                     GETEDO2(x14);
                     emit_shr32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 7:
                     INST_NAME("SAR Ed, 1");
                     SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                     GETEDO2(x14);
                     emit_sar32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
             }
             break;
@@ -546,7 +552,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     RSB_IMM8(x3, x3, 0x20);
                     GETEDO2(x14);
                     MOV_REG_ROR_REG(ed, ed, x3);
-                    WBACK;
+                    WBACK2;
                     UFLAG_IF {  // calculate flags directly
                         CMPS_IMM8(x3, 31);
                         B_MARK(cNE);
@@ -567,7 +573,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     B_MARK2(cEQ);
                     GETEDO2(x14);
                     MOV_REG_ROR_REG(ed, ed, x3);
-                    WBACK;
+                    WBACK2;
                     UFLAG_IF {  // calculate flags directly
                         CMPS_IMM8(x3, 1);
                         B_MARK(cNE);
@@ -587,10 +593,9 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     READFLAGS(X_CF);
                     SETFLAGS(X_OF|X_CF, SF_SET);
                     AND_IMM8(x2, xECX, 0x1f);
-                    wb = ed;
                     if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
-                    CALL_(rcl32, ed, (1<<x14));
-                    SBACK(wb);
+                    CALL_(rcl32, ed, (1<<x2));
+                    WBACK2;
                     break;
                 case 3:
                     INST_NAME("RCR Ed, CL");
@@ -598,10 +603,9 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     READFLAGS(X_CF);
                     SETFLAGS(X_OF|X_CF, SF_SET);
                     AND_IMM8(x2, xECX, 0x1f);
-                    wb = ed;
                     if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
                     CALL_(rcr32, ed, (1<<x14));
-                    SBACK(wb);
+                    WBACK2;
                     break;
                 case 4:
                 case 6:
@@ -610,7 +614,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     AND_IMM8(x3, xECX, 0x1f);
                     GETEDO2(x14);
                     emit_shl32(dyn, ninst, ed, x3, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 5:
                     INST_NAME("SHR Ed, CL");
@@ -618,7 +622,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     AND_IMM8(x3, xECX, 0x1f);
                     GETEDO2(x14);
                     emit_shr32(dyn, ninst, ed, x3, x3, x14);
-                    WBACK;
+                    WBACK2;
                     break;
                 case 7:
                     INST_NAME("SAR Ed, CL");
@@ -627,7 +631,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     GETEDO2(x14);
                     UFLAG_OP12(ed, x3);
                     MOV_REG_ASR_REG(ed, ed, x3);
-                    WBACK;
+                    WBACK2;
                     UFLAG_RES(ed);
                     UFLAG_DF(x3, d_sar32);
                     break;
@@ -649,6 +653,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {   // reg
                         DEFAULT;
                     } else {                    // mem <= i32
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                         LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                         PUSH1(x3);

src/dynarec/dynarec_arm_66.c

@@ -505,7 +505,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
 
         case 0x87:
-            INST_NAME("XCHG Ew, Gw");
+            INST_NAME("(LOCK) XCHG Ew, Gw");
             nextop = F8;
             if((nextop&0xC0)==0xC0) {
                 GETGD;
@@ -518,6 +518,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {
                 GETGD;
                 UXTH(x14, gd, 0);
+                SMDMB();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, LOCK_LOCK);
                 #if 0
                 MARKLOCK;
@@ -531,6 +532,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 LDRH_IMM8(x1, ed, fixedaddress);
                 STRH_IMM8(x14, ed, fixedaddress);
                 #endif
+                SMDMB();
                 BFI(gd, x1, 0, 16);
             }
             break;
@@ -547,7 +549,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, &lock);
                 STRH_IMM8(gd, ed, fixedaddress);
-                if(lock) {DMB_ISH();}
+                SMWRITELOCK(lock);
             }
             break;
         case 0x8B:
@@ -561,7 +563,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 }
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, &lock);
-                if(lock) {DMB_ISH();}
+                SMREADLOCK(lock);
                 LDRH_IMM8(x1, ed, fixedaddress);
                 BFI(gd, x1, 0, 16);
             }
@@ -577,6 +579,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 STRH_IMM8(x1, ed, fixedaddress);
+                SMWRITE2();
             }
             break;
 
@@ -588,6 +591,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
                 STRH_REG(ed, xEmu, x2);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 LDRH_IMM8(x1, ed, fixedaddress);
                 MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
@@ -645,6 +649,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             INST_NAME("MOV, AX, Od");
             u32 = F32;
             MOV32(x2, u32);
+            SMREAD();
             LDRH_IMM8(x2, x2, 0);
             BFI(xEAX, x2, 0, 16);
             break;
@@ -653,6 +658,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             u32 = F32;
             MOV32(x2, u32);
             STRH_IMM8(xEAX, x2, 0);
+            SMWRITE();
             break;
         case 0xA5:
             INST_NAME("MOVSW");

src/dynarec/dynarec_arm_660f.c

@@ -27,9 +27,10 @@
     if((nextop&0xC0)==0xC0) {   \
         a = sse_get_reg(dyn, ninst, x1, nextop&7, w);  \
     } else {                    \
+        SMREAD();               \
         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL); \
         a = fpu_get_scratch_quad(dyn); \
-        VLD1Q_8(a, ed);       \
+        VLD1Q_8(a, ed);         \
     }
 #define GETGX(a, w)         \
     gd = (nextop&0x38)>>3;  \
@@ -70,6 +71,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
@@ -96,6 +98,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 VMOVfrV_D(x2, x3, v0+1);
                 STR_IMM9(x2, ed, fixedaddress+8);
                 STR_IMM9(x3, ed, fixedaddress+12);
+                SMWRITE2();
             }
             break;
         case 0x12:
@@ -108,6 +111,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 return addr;
             }
             parity = getedparity(dyn, ninst, addr, nextop, 3);
+            SMREAD();
             if(parity) {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 VLDR_64(v0, ed, fixedaddress);
@@ -137,6 +141,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 STR_IMM9(x2, ed, fixedaddress);
                 STR_IMM9(x3, ed, fixedaddress+4);
             }
+            SMWRITE2();
             break;
         case 0x14:
             INST_NAME("UNPCKLPD Gx, Ex");
@@ -146,6 +151,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 VMOVD(v0+1, v1);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 VLDR_64(v0+1, ed, fixedaddress);
             }
@@ -159,6 +165,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 VMOVD(v0+1, v1+1);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023-4, 3, 0, NULL);
                 VLDR_64(v0+1, ed, fixedaddress+4);
             }
@@ -173,6 +180,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 return addr;
             }
             parity = getedparity(dyn, ninst, addr, nextop, 3);
+            SMREAD();
             if(parity) {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 VLDR_64(v0+1, ed, fixedaddress);    // vfpu opcode here
@@ -202,6 +210,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 STR_IMM9(x2, ed, fixedaddress);
                 STR_IMM9(x3, ed, fixedaddress+4);
             }
+            SMWRITE2();
             break;
 
         case 0x1F:
@@ -221,7 +230,8 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
-                VLD1Q_32(v0, ed);
+                VLD1Q_64(v0, ed);
+                SMWRITE2();
             }
             break;
         case 0x29:
@@ -234,7 +244,24 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 VMOVQ(v1, v0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
-                VST1Q_32(v0, ed);
+                VST1Q_64(v0, ed);
+                SMWRITE2();
+            }
+            break;
+
+
+        case 0x2B:
+            INST_NAME("MOVNTPD Ex, Gx");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
+            if((nextop&0xC0)==0xC0) {
+                v1 = sse_get_reg_empty(dyn, ninst, x1, nextop&7);
+                VMOVQ(v1, v0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                VST1Q_64(v0, ed);
+                SMWRITE2();
             }
             break;
 
@@ -262,6 +289,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         v0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         VLD1Q_64(q1, ed);
                         v0 = q1;
@@ -469,6 +497,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             if((nextop&0xC0)==0xC0) {   \
                 ed = xEAX+(nextop&7);   \
             } else { \
+                SMREAD(); \
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);\
                 LDRH_IMM8_COND(YES, x1, ed, fixedaddress); \
                 ed = x1;                        \
@@ -687,8 +716,26 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             nextop = F8;
             GETGX(v0, 1);
             GETEX(v1, 0);
-            // need rounding?
-            VCVTQ_S32_F32(v0, v1);
+            if(v0>15)
+                q0 = fpu_get_scratch_quad(dyn);
+            else
+                q0 = v0;
+            if(v1>15) {
+                q1 = fpu_get_scratch_quad(dyn);
+                VMOVQ(q1, v1);
+            } else
+                q1 = v1;
+            // need rounding!
+            //VCVTQ_S32_F32(v0, v1);
+            u8 = sse_setround(dyn, ninst, x1, x2, x14);
+            VCVTR_S32_F32(q0*2+0, q1*2+0);
+            VCVTR_S32_F32(q0*2+1, q1*2+1);
+            VCVTR_S32_F32(q0*2+2, q1*2+2);
+            VCVTR_S32_F32(q0*2+3, q1*2+3);
+            if(q0!=v0) {
+                VMOVQ(v0, q0);
+            }
+            x87_restoreround(dyn, ninst, u8);
             break;
         case 0x5C:
             INST_NAME("SUBPD Gx, Ex");
@@ -893,6 +940,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 VMOVD(v0+1, v1);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 VLDR_64(v0+1, ed, fixedaddress);
             }
@@ -906,6 +954,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 VMOVD(q0+1, q1+1);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023-8, 3, 0, NULL);
                 VLDR_64(q0+1, ed, fixedaddress+8);
             }
@@ -928,6 +977,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 VLD1Q_64(v0, ed);
@@ -1043,6 +1093,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     VMOVQ(v0, q0);
                 }
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 u8 = F8;
@@ -1071,6 +1122,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_16(q0, ed);
@@ -1084,6 +1136,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1Q_16(q0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1092,6 +1145,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_16(q0, ed);
@@ -1102,6 +1156,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     }
                     if((nextop&0xC0)!=0xC0) {
                         VST1Q_16(q0, ed);
+                        SMWRITE2();
                     }
                     break;
                 case 6:
@@ -1109,6 +1164,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_16(q0, ed);
@@ -1122,6 +1178,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1Q_16(q0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1138,6 +1195,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_32(q0, ed);
@@ -1151,6 +1209,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1Q_32(q0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1159,6 +1218,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_32(q0, ed);
@@ -1169,6 +1229,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     }
                     if((nextop&0xC0)!=0xC0) {
                         VST1Q_32(q0, ed);
+                        SMWRITE2();
                     }
                     break;
                 case 6:
@@ -1176,6 +1237,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_32(q0, ed);
@@ -1189,6 +1251,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1Q_32(q0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1204,6 +1267,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_64(q0, ed);
@@ -1217,6 +1281,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1Q_64(q0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1225,6 +1290,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_64(q0, ed);
@@ -1240,6 +1306,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1Q_64(q0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1248,6 +1315,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_64(q0, ed);
@@ -1261,6 +1329,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1Q_64(q0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1269,6 +1338,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         q0 = fpu_get_scratch_quad(dyn);
                         VLD1Q_64(q0, ed);
@@ -1284,6 +1354,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         }
                         if((nextop&0xC0)!=0xC0) {
                             VST1Q_64(q0, ed);
+                            SMWRITE2();
                         }
                     }
                     break;
@@ -1325,6 +1396,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 VMOVfrDx_32(x2, v0, 0); // to avoid Bus Error, using regular store
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0, NULL);
                 STR_IMM9(x2, ed, fixedaddress);
+                SMWRITE2();
             }
             break;
         case 0x7F:
@@ -1337,6 +1409,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 VST1Q_32(v0, ed);
+                SMWRITE2();
             }
             break;
 
@@ -1349,6 +1422,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);   // no need for extract
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, 255, 0, 0, NULL);
                 SBFX(x1, gd, 4, 12);    // r1 = (gw>>4);
                 ADD_REG_LSL_IMM5(x1, ed, x1, 1); //(&ed)+=r1*2;
@@ -1391,6 +1465,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 ed = xEAX+(nextop&7);
                 wback = 0;   // no need for extract
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                 SBFX(x14, gd, 4, 12);    // r14 = (gw>>4);
                 ADD_REG_LSL_IMM5(x3, wback, x14, 1); //(&ew)+=r14*2;
@@ -1407,6 +1482,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             ORR_REG_LSL_REG(ed, ed, x1, x2);
             if(wback) {
                 STRH_IMM8(ed, wback, fixedaddress);
+                SMWRITE();
             }
             break;
         case 0xAC:
@@ -1459,6 +1535,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 ed = xEAX+(nextop&7);
                 wback = 0;   // no need for extract
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                 SBFX(x14, gd, 4, 12);    // r14 = (gw>>4);
                 ADD_REG_LSL_IMM5(x3, wback, x14, 1); //(&ew)+=r14*2;
@@ -1475,6 +1552,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             XOR_REG_LSL_REG(ed, ed, x1, x2);
             if(wback) {
                 STRH_IMM8(ed, wback, fixedaddress);
+                SMWRITE();
             }
             break;
 
@@ -1488,6 +1566,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 eb2 = (ed&4)>>2;    // L or H
                 UXTB(x1, eb1, eb2);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 4095, 0, 0, NULL);
                 LDRB_IMM9(x1, ed, fixedaddress);
             }
@@ -1501,6 +1580,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 ed = xEAX + (nextop&7);
                 UXTH(x1, ed, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 LDRH_IMM8(x1, ed, fixedaddress);
             }
@@ -1517,6 +1597,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                         LDRH_IMM8(x1, wback, fixedaddress);
                         ed = x1;
@@ -1537,6 +1618,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         ed = xEAX+(nextop&7);
                         wback = 0;
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                         LDRH_IMM8(x1, wback, fixedaddress);
                         ed = x1;
@@ -1555,6 +1637,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     XOR_REG_LSL_IMM5(ed, ed, x14, u8);
                     if(wback) {
                         STRH_IMM8(ed, wback, fixedaddress);
+                        SMWRITE();
                     }
                     MARK3;
                     break;
@@ -1566,6 +1649,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         ed = xEAX+(nextop&7);
                         wback = 0;
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                         LDRH_IMM8(x1, wback, fixedaddress);
                         ed = x1;
@@ -1584,6 +1668,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     XOR_REG_LSL_IMM5(ed, ed, x14, u8);
                     if(wback) {
                         STRH_IMM8(ed, wback, fixedaddress);
+                        SMWRITE();
                     }
                     MARK3;
                     break;
@@ -1595,6 +1680,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         ed = xEAX+(nextop&7);
                         wback = 0;
                     } else {
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                         LDRH_IMM8(x1, wback, fixedaddress);
                         ed = x1;
@@ -1612,6 +1698,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     XOR_REG_LSL_IMM5(ed, ed, x14, u8);
                     if(wback) {
                         STRH_IMM8(ed, wback, fixedaddress);
+                        SMWRITE();
                     }
                     MARK3;
                     break;
@@ -1629,6 +1716,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 ed = xEAX+(nextop&7);
                 wback = 0;   // no need for extract
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                 SBFX(x14, gd, 4, 12);    // r14 = (gw>>4);
                 ADD_REG_LSL_IMM5(x3, wback, x14, 1); //(&ew)+=r14*2;
@@ -1643,6 +1731,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             XOR_REG_LSL_REG(ed, ed, x1, x2);
             if(wback) {
                 STRH_IMM8(ed, wback, fixedaddress);
+                SMWRITE();
             }
             break;
         case 0xBC:
@@ -1688,6 +1777,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 eb2 = (ed&4)>>2;    // L or H
                 SXTB(x1, eb1, eb2);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 LDRSB_IMM8(x1, ed, fixedaddress);
             }
@@ -1759,6 +1849,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 ed = xEAX+(nextop&7);
                 VMOVtoDx_16(v0+(u8/4), u8&3, ed);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0, NULL);
                 u8 = (F8)&7;
                 VLD1LANE_16(v0+(u8/4), wback, u8&3);
@@ -1773,6 +1864,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 u8 = (F8)&7;
                 VMOVfrDx_U16(gd, v0+(u8/4), (u8&3));
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255-14, 0, 0, NULL);
                 u8 = (F8)&7;
                 LDRH_IMM8(gd, wback, fixedaddress+u8*2);
@@ -1830,8 +1922,11 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGX(q0, 1);
             GETEX(q1, 0);
             v0 = fpu_get_scratch_quad(dyn);
-            VDUPQ_16(v0, q1, 0);
-            VNEGNQ_16(v0, v0);   // because we want SHR and not SHL
+            VQMOVN_S64(v0, q1);
+            VQMOVN_S32(v0, v0);
+            VQMOVN_S16(v0, v0);
+            VDUPQ_8(v0, v0, 0); // only 8 bits are taken
+            VNEGNQ_8(v0, v0);   // because we want SHR and not SHL
             VSHLQ_U16(q0, q0, v0);
             break;
         case 0xD2:
@@ -1840,11 +1935,11 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGX(q0, 1);
             GETEX(q1, 0);
             v0 = fpu_get_scratch_quad(dyn);
-            VMOVD(v0, q1);
-            VMOVD(v0+1, q1);
-            VQMOVN_S64(v0, v0); // 2*q1 in 32bits now
-            VNEGN_32(v0, v0);   // because we want SHR and not SHL
-            VMOVD(v0+1, v0);
+            VQMOVN_S64(v0, q1);
+            VQMOVN_S32(v0, v0);
+            VQMOVN_S16(v0, v0);
+            VDUPQ_8(v0, v0, 0); // only 8 bits are taken
+            VNEGNQ_8(v0, v0);   // because we want SHR and not SHL
             VSHLQ_U32(q0, q0, v0);
             break;
         case 0xD3:
@@ -1853,11 +1948,11 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGX(q0, 1);
             GETEX(q1, 0);
             v0 = fpu_get_scratch_quad(dyn);
-            VMOVD(v0, q1);
-            VMOVD(v0+1, q1);
-            VQMOVN_S64(v0, v0); // 2*q1 in 32bits now
-            VNEGN_32(v0, v0);   // because we want SHR and not SHL
-            VMOVL_S32(v0, v0);  // 2*q1 in 64 bits now
+            VQMOVN_S64(v0, q1);
+            VQMOVN_S32(v0, v0);
+            VQMOVN_S16(v0, v0);
+            VDUPQ_8(v0, v0, 0); // only 8 bits are taken
+            VNEGNQ_8(v0, v0);   // because we want SHR and not SHL
             VSHLQ_U64(q0, q0, v0);
             break;
         case 0xD4:
@@ -1894,6 +1989,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     STR_IMM9(x2, ed, fixedaddress+0);
                     STR_IMM9(x3, ed, fixedaddress+4);
                 }
+                SMWRITE2();
             }
             break;
         case 0xD7:
@@ -1999,13 +2095,11 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGX(q0, 1);
             GETEX(q1, 0);
             v0 = fpu_get_scratch_quad(dyn);
-            VMOVD(v0, q1);
-            VMOVD(v0+1, q1);
-            VQMOVN_S64(v0, v0); // 2*q1 in 32bits now
-            VMOVD(v0+1, v0);
-            VQMOVN_S32(v0, v0); // 4*q1 in 16bits now
-            VNEGN_16(v0, v0);   // because we want SHR and not SHL
-            VMOVD(v0+1, v0);
+            VQMOVN_S64(v0, q1);
+            VQMOVN_S32(v0, v0);
+            VQMOVN_S16(v0, v0);
+            VDUPQ_8(v0, v0, 0); // only 8 bits are taken
+            VNEGNQ_8(v0, v0);   // because we want SHR and not SHL
             VSHLQ_S16(q0, q0, v0);
             break;
         case 0xE2:
@@ -2014,11 +2108,11 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGX(q0, 1);
             GETEX(q1, 0);
             v0 = fpu_get_scratch_quad(dyn);
-            VMOVD(v0, q1);
-            VMOVD(v0+1, q1);
-            VQMOVN_S64(v0, v0); // 2*q1 in 32bits now
-            VNEGN_32(v0, v0);   // because we want SHR and not SHL
-            VMOVD(v0+1, v0);
+            VQMOVN_S64(v0, q1);
+            VQMOVN_S32(v0, v0);
+            VQMOVN_S16(v0, v0);
+            VDUPQ_8(v0, v0, 0); // only 8 bits are taken
+            VNEGNQ_8(v0, v0);   // because we want SHR and not SHL
             VSHLQ_S32(q0, q0, v0);
             break;
         case 0xE3:
@@ -2148,12 +2242,10 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGX(q0, 1);
             GETEX(q1, 0);
             v0 = fpu_get_scratch_quad(dyn);
-            VMOVD(v0, q1);
-            VMOVD(v0+1, q1);
-            VQMOVN_S64(v0, v0); // 2*q1 in 32bits now
-            VMOVD(v0+1, v0);
-            VQMOVN_S32(v0, v0); // 4*q1 in 16bits now
-            VMOVD(v0+1, v0);
+            VQMOVN_S64(v0, q1);
+            VQMOVN_S32(v0, v0);
+            VQMOVN_S16(v0, v0);
+            VDUPQ_8(v0, v0, 0); // only 8 bits are taken
             VSHLQ_U16(q0, q0, v0);
             break;
         case 0xF2:
@@ -2162,10 +2254,10 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGX(q0, 1);
             GETEX(q1, 0);
             v0 = fpu_get_scratch_quad(dyn);
-            VMOVD(v0, q1);
-            VMOVD(v0+1, q1);
-            VQMOVN_S64(v0, v0); // 2*q1 in 32bits now
-            VMOVD(v0+1, v0);
+            VQMOVN_S64(v0, q1);
+            VQMOVN_S32(v0, v0);
+            VQMOVN_S16(v0, v0);
+            VDUPQ_8(v0, v0, 0); // only 8 bits are taken
             VSHLQ_U32(q0, q0, v0);
             break;
         case 0xF3:
@@ -2174,8 +2266,10 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGX(q0, 1);
             GETEX(q1, 0);
             v0 = fpu_get_scratch_quad(dyn);
-            VMOVD(v0, q1);
-            VMOVD(v0+1, q1);
+            VQMOVN_S64(v0, q1);
+            VQMOVN_S32(v0, v0);
+            VQMOVN_S16(v0, v0);
+            VDUPQ_8(v0, v0, 0); // only 8 bits are taken
             VSHLQ_U64(q0, q0, v0);
             break;
         case 0xF4:
@@ -2217,7 +2311,7 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             VPADDLQ_U32(q0, q0);
             break;
         case 0xF7:
-            INST_NAME("MASKMOVDQU Gx, Ex")
+            INST_NAME("MASKMOVDQU Gx, Ex");
             nextop = F8;
             GETGX(q0, 1);
             GETEX(q1, 0);

src/dynarec/dynarec_arm_66f0.c

@@ -41,7 +41,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
         case 0x81:
         case 0x83:
             nextop = F8;
-            DMB_ISH();
+            SMDMB();
             switch((nextop>>3)&7) {
                 case 0: //ADD
                     if(opcode==0x81) {
@@ -50,7 +50,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                         INST_NAME("LOCK ADD Ew, Ib");
                     }
                     SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         if(opcode==0x81) i16 = F16S; else i16 = F8S;
                         ed = xEAX+(nextop&7);
@@ -71,7 +71,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                             B_MARKLOCK(cNE);
                         }
                         if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                             B_NEXT(c__);
                         }
                         if(!fixedaddress || (fixedaddress && (fixedaddress&1))) {
@@ -85,17 +85,17 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                             B_MARK(cNE);
                             STRH_IMM8(x1, wback, 0); // put the whole value
                         }
-                        DMB_ISH();
+                        SMDMB();
                     }
                     break;
                 default:
                   DEFAULT;
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0xFF:
             nextop = F8;
-            DMB_ISH();
+            SMDMB();
             switch((nextop>>3)&7)
             {
                 case 0: // INC Ew
@@ -120,7 +120,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                             B_MARKLOCK(cNE);
                         }
                         if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                             B_NEXT(c__);
                         }
                         if(!fixedaddress || (fixedaddress && (fixedaddress&1))) {
@@ -158,7 +158,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                             B_MARKLOCK(cNE);
                         }
                         if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                             B_NEXT(c__);
                         }
                         if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -178,7 +178,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     // default to NO LOCK
                     addr-=2;
             }
-            DMB_ISH();
+            SMDMB();
             break;
         default:
             DEFAULT;

src/dynarec/dynarec_arm_67.c

@@ -58,6 +58,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     } else {
                         addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                         STRH_REG(gd, x14, ed);
+                        SMWRITE2();
                     }
                     break;
 
@@ -72,6 +73,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             BFI(gd, ed, 0, 16);
                         }
                     } else {                    // mem <= reg
+                        SMREAD();
                         addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                         LDR_REG_LSL_IMM5(gd, x14, ed, 0);
                     }
@@ -87,6 +89,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         addr = geted16(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
                         POP1(x2);
                         STR_REG_LSL_IMM5(x2, x14, ed, 0);
+                        SMWRITE2();
                     }
                     break;
 
@@ -98,6 +101,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         MOV32(x2, u32);
                         ADD_REG_LSL_IMM5(x1, x1, x2, 0);
                     }
+                    SMREAD();
                     LDR_IMM9(xEAX, x1, 0);
                     break;
 
@@ -110,6 +114,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         ADD_REG_LSL_IMM5(x1, x1, x2, 0);
                     }
                     STR_IMM9(xEAX, x1, 0);
+                    SMWRITE2();
                     break;
 
                 case 0xFF:
@@ -121,6 +126,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             if((nextop&0xC0)==0xC0) {   // reg
                                 DEFAULT;
                             } else {                    // mem <= i32
+                                SMREAD();
                                 addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                                 LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                                 PUSH1(x3);

src/dynarec/dynarec_arm_d9.c

@@ -194,7 +194,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
         case 0xE8:
             INST_NAME("FLD1");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 VMOV_i_32(v1, 0b01110000);
             } else {
@@ -204,7 +204,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xE9:
             INST_NAME("FLDL2T");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 MOV32(x2, (&f_l2t));
                 VLDR_32(v1, x2, 0);
@@ -216,7 +216,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xEA:     
             INST_NAME("FLDL2E");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 MOV32(x2, (&f_l2e));
                 VLDR_32(v1, x2, 0);
@@ -228,7 +228,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xEB:
             INST_NAME("FLDPI");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 MOV32(x2, (&f_pi));
                 VLDR_32(v1, x2, 0);
@@ -240,7 +240,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xEC:
             INST_NAME("FLDLG2");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 MOV32(x2, (&f_lg2));
                 VLDR_32(v1, x2, 0);
@@ -252,7 +252,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xED:
             INST_NAME("FLDLN2");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 MOV32(x2, (&f_ln2));
                 VLDR_32(v1, x2, 0);
@@ -264,7 +264,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xEE:
             INST_NAME("FLDZ");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 VMOV_8(v1/2, 0);  // float is *2...
             } else {
@@ -309,7 +309,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xF2:
             INST_NAME("FPTAN");
-            v2 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            v2 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
             v1 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
             VMOV_64(0, v1);    // prepare call to tan
             CALL_1D(tan, 0);
@@ -589,7 +589,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 2:
                     INST_NAME("FST float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                     if(ST_IS_F(0))
                         s0 = v1;
                     else {
@@ -608,7 +608,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 3:
                     INST_NAME("FSTP float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                     if(ST_IS_F(0))
                         s0 = v1;
                     else {

src/dynarec/dynarec_arm_db.c

@@ -223,7 +223,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 case 2:
                     INST_NAME("FIST Ed, ST0");
                     v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x14); // x1 have the modified RPSCR reg
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14); // x1 have the modified RPSCR reg
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                         wback = 0;
@@ -233,10 +233,6 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     }
                     s0 = fpu_get_scratch_single(dyn);
                     MSR_nzcvq_0();
-                    // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                     VCVTR_S32_F64(s0, v1);
                     VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
                     VMOVfrV(ed, s0);
@@ -248,7 +244,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 case 3:
                     INST_NAME("FISTP Ed, ST0");
                     v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x14); // x1 have the modified RPSCR reg
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14); // x1 have the modified RPSCR reg
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                         wback = 0;
@@ -258,10 +254,6 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     }
                     s0 = fpu_get_scratch_single(dyn);
                     MSR_nzcvq_0();
-                    // The FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                     VCVTR_S32_F64(s0, v1);
                     VMRS(x3);   // get the FPCSR reg and test FPU execption (denormal and overflow only)
                     VMOVfrV(ed, s0);
@@ -287,17 +279,89 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         STR_IMM9(x2, ed, 0);
                         STR_IMM9(x3, ed, 4);
                         STRH_IMM8(x14, ed, 8);
+                    } else {
+                        if(box86_x87_no80bits) {
+                            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+                            parity = getedparity(dyn, ninst, addr, nextop, 3);
+                            if (parity) {
+                                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                                VLDR_64(v1, ed, fixedaddress);
+                            } else {
+                                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
+                                LDR_IMM9(x2, ed, fixedaddress);
+                                LDR_IMM9(x3, ed, fixedaddress + 4);
+                                VMOVtoV_D(v1, x2, x3);
+                            }
+                        } else {
+                            #if 0
+                            MESSAGE(LOG_DUMP, "Need Optimization\n");
+                            if(ed!=x1) {
+                                MOV_REG(x1, ed);
+                            }
+                            x87_do_push_empty(dyn, ninst, x3);
+                            CALL(arm_fld, -1, 0);
+                            #else
+                            v1 = x87_do_push(dyn, ninst, x2, NEON_CACHE_ST_D);
+                            // copy 10bytes of *ED to STld(0)
+                            LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
+                            int a = -dyn->n.x87stack;
+                            if(a) {
+                                if(a<0) {
+                                    SUB_IMM8(x3, x3, -a);
+                                } else {
+                                    ADD_IMM8(x3, x3, a);
+                                }
+                                AND_IMM8(x3, x3, 7);
+                            }
+                            ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
+                            MOVW(x14, sizeof(fpu_ld_t));
+                            MLA(x2, x3, x14, x2);
+                            LDR_IMM9(x14, ed, 0);
+                            STR_IMM9(x14, x2, 0);
+                            LDR_IMM9(x14, ed, 4);
+                            STR_IMM9(x14, x2, 4);
+                            LDRH_IMM8(x14, ed, 8);
+                            STRH_IMM8(x14, x2, 8);
+                            // convert 10bytes -> double
+                            CALL_1RD(FromLD, ed, x2, (1<<x3)|(1<<xFlags)|(1<<x2));
+                            // store ref to STld(0).uref, but can be unaligned...
+                            VMOVfrV_D(x3, x14, 0);
+                            // also, store result to ST0
+                            VMOVD(v1, 0);
+                            STR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
+                            STR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                            #endif
+                        }
+                    }
+                    break;
+                case 7:
+                    INST_NAME("FSTP tbyte");
+                        if(box86_x87_no80bits) {
+                        v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                        parity = getedparity(dyn, ninst, addr, nextop, 3);
+                        if(parity) {
+                            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                            VSTR_64(v1, ed, fixedaddress);
+                        } else {
+                            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
+                            VMOVfrV_D(x2, x3, v1);
+                            STR_IMM9(x2, ed, fixedaddress);
+                            STR_IMM9(x3, ed, fixedaddress+4);
+                        }
                     } else {
                         #if 0
                         MESSAGE(LOG_DUMP, "Need Optimization\n");
+                        x87_forget(dyn, ninst, x1, x3, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                         if(ed!=x1) {
                             MOV_REG(x1, ed);
                         }
-                        x87_do_push_empty(dyn, ninst, x3);
-                        CALL(arm_fld, -1, 0);
+                        CALL(arm_fstp, -1, 0);
                         #else
-                        v1 = x87_do_push(dyn, ninst, x2, NEON_CACHE_ST_D);
-                        // copy 10bytes of *ED to STld(0)
+                        v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                        v2 = fpu_get_scratch_double(dyn);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                        // get top in x3
                         LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
                         int a = -dyn->n.x87stack;
                         if(a) {
@@ -308,85 +372,41 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             }
                             AND_IMM8(x3, x3, 7);
                         }
+                        // get STld(0)
                         ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
                         MOVW(x14, sizeof(fpu_ld_t));
                         MLA(x2, x3, x14, x2);
-                        LDR_IMM9(x14, ed, 0);
-                        STR_IMM9(x14, x2, 0);
-                        LDR_IMM9(x14, ed, 4);
-                        STR_IMM9(x14, x2, 4);
-                        LDRH_IMM8(x14, ed, 8);
-                        STRH_IMM8(x14, x2, 8);
-                        // convert 10bytes -> double
-                        CALL_1RD(FromLD, ed, x2, (1<<x3)|(1<<xFlags)|(1<<x2));
-                        // store ref to STld(0).uref, but can be unaligned...
-                        VMOVfrV_D(x3, x14, 0);
-                        // also, store result to ST0
-                        VMOVD(v1, 0);
-                        STR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
-                        STR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                        // compare uref with actual value, top is not used anymore, so x3 is free
+                        LDR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
+                        LDR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                        VMOVtoV_D(v2, x14, x3);
+                        VCEQ_32(v2, v2, v1);    // same?
+                        VMOVfrV_D(x14, x3, v2);
+                        TSTS_REG_LSL_IMM5(x14, x3, 0);
+                        B_MARK(cEQ);    // different, do the conversion
+                        // same... copy STld(0).ld
+                        LDR_IMM9(x14, x2, 0);
+                        LDR_IMM9(x3, x2, 4);
+                        STR_IMM9(x14, ed, 0);
+                        STR_IMM9(x3, ed, 4);
+                        LDRH_IMM8(x14, x2, 8);
+                        STRH_IMM8(x14, ed, 8);
+                        B_MARK2(c__);
+                        MARK;
+                        // different, call D2LD
+                        SUB_IMM8(xSP, xSP, 8);
+                        VSTR_64(v1, xSP, 0);
+                        PUSH(xSP, (1<<xEmu)|(1<<xFlags));
+                        ADD_IMM8(0, xSP, 8);    //1st arg is &d
+                        if(ed!=x1) {
+                            MOV_REG_LSL_IMM5(x1, ed, 0);    //2nd arg is ed
+                        }
+                        CALL_S(D2LD, -2, 0);
+                        POP(xSP, (1<<xEmu)|(1<<xFlags));
+                        ADD_IMM8(xSP, xSP, 8);
+                        MARK2;
                         #endif
                     }
-                    break;
-                case 7:
-                    INST_NAME("FSTP tbyte");
-                    #if 0
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    x87_forget(dyn, ninst, x1, x3, 0);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
-                    if(ed!=x1) {
-                        MOV_REG(x1, ed);
-                    }
-                    CALL(arm_fstp, -1, 0);
-                    #else
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch_double(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
-                    // get top in x3
-                    LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
-                    int a = -dyn->n.x87stack;
-                    if(a) {
-                        if(a<0) {
-                            SUB_IMM8(x3, x3, -a);
-                        } else {
-                            ADD_IMM8(x3, x3, a);
-                        }
-                        AND_IMM8(x3, x3, 7);
-                    }
-                    // get STld(0)
-                    ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
-                    MOVW(x14, sizeof(fpu_ld_t));
-                    MLA(x2, x3, x14, x2);
-                    // compare uref with actual value, top is not used anymore, so x3 is free
-                    LDR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
-                    LDR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
-                    VMOVtoV_D(v2, x14, x3);
-                    VCEQ_32(v2, v2, v1);    // same?
-                    VMOVfrV_D(x14, x3, v2);
-                    TSTS_REG_LSL_IMM5(x14, x3, 0);
-                    B_MARK(cEQ);    // different, do the conversion
-                    // same... copy STld(0).ld
-                    LDR_IMM9(x14, x2, 0);
-                    LDR_IMM9(x3, x2, 4);
-                    STR_IMM9(x14, ed, 0);
-                    STR_IMM9(x3, ed, 4);
-                    LDRH_IMM8(x14, x2, 8);
-                    STRH_IMM8(x14, ed, 8);
-                    B_MARK2(c__);
-                    MARK;
-                    // different, call D2LD
-                    SUB_IMM8(xSP, xSP, 8);
-                    VSTR_64(v1, xSP, 0);
-                    PUSH(xSP, (1<<xEmu)|(1<<xFlags));
-                    ADD_IMM8(0, xSP, 8);    //1st arg is &d
-                    if(ed!=x1) {
-                        MOV_REG_LSL_IMM5(x1, ed, 0);    //2nd arg is ed
-                    }
-                    CALL_S(D2LD, -2, 0);
-                    POP(xSP, (1<<xEmu)|(1<<xFlags));
-                    ADD_IMM8(xSP, xSP, 8);
-                    MARK2;
-                    #endif
                     x87_do_pop(dyn, ninst, x3);
                     break;
                 default:

src/dynarec/dynarec_arm_df.c

@@ -161,7 +161,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FILD ST0, Ew");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+                    v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                     LDRSH_IMM8(x1, wback, fixedaddress);
                     if(ST_IS_F(0)) {
@@ -175,7 +175,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 1:
                     INST_NAME("FISTTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
@@ -202,16 +202,12 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 2:
                     INST_NAME("FIST Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x14);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
                     MSR_nzcvq_0();
-                    // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                     if(ST_IS_F(0)) {
                         VCVTR_S32_F32(s0, v1);
                     } else {
@@ -228,16 +224,12 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 3:
                     INST_NAME("FISTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x14);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
                     MSR_nzcvq_0();
-                    // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                     if(ST_IS_F(0)) {
                         VCVTR_S32_F32(s0, v1);
                     } else {

src/dynarec/dynarec_arm_f0.c

@@ -42,7 +42,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGB(x2);
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 wback = (nextop&7);    
                 wb2 = (wback>>2);      
@@ -59,14 +59,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x14, 0);
                 B_MARKLOCK(cNE);    // write failed, try again
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x01:
             INST_NAME("LOCK ADD Ed, Gd");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGD;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);
                 emit_add32(dyn, ninst, ed, gd, x3, x14);
@@ -79,13 +79,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x3, 0);
                 B_MARKLOCK(cNE);
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x08:
             INST_NAME("LOCK OR Eb, Gb");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 GETGB(x2);
                 wback = (nextop&7);    
@@ -104,14 +104,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x14, 0);
                 B_MARKLOCK(cNE);    // write failed, try again
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x09:
             INST_NAME("LOCK OR Ed, Gd");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGD;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);
                 emit_or32(dyn, ninst, ed, gd, x3, x14);
@@ -124,7 +124,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x3, 0);
                 B_MARKLOCK(cNE);
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x10:
             INST_NAME("LOCK ADC Eb, Gb");
@@ -132,7 +132,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGB(x2);
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 wback = (nextop&7);    
                 wb2 = (wback>>2);      
@@ -149,7 +149,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x14, 0);
                 B_MARKLOCK(cNE);    // write failed, try again
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x11:
             INST_NAME("LOCK ADC Ed, Gd");
@@ -157,7 +157,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGD;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);
                 emit_adc32(dyn, ninst, ed, gd, x3, x14);
@@ -170,7 +170,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x3, 0);
                 B_MARKLOCK(cNE);
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x18:
             INST_NAME("LOCK SBB Eb, Gb");
@@ -178,7 +178,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGB(x2);
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 wback = (nextop&7);    
                 wb2 = (wback>>2);      
@@ -195,7 +195,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x14, 0);
                 B_MARKLOCK(cNE);    // write failed, try again
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x19:
             INST_NAME("LOCK SBB Ed, Gd");
@@ -203,7 +203,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGD;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);
                 emit_sbb32(dyn, ninst, ed, gd, x3, x14);
@@ -216,13 +216,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x3, 0);
                 B_MARKLOCK(cNE);
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x20:
             INST_NAME("LOCK AND Eb, Gb");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 GETGB(x2);
                 wback = (nextop&7);    
@@ -241,14 +241,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x14, 0);
                 B_MARKLOCK(cNE);    // write failed, try again
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x21:
             INST_NAME("LOCK AND Ed, Gd");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGD;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);
                 emit_and32(dyn, ninst, ed, gd, x3, x14);
@@ -261,14 +261,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x3, 0);
                 B_MARKLOCK(cNE);
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x28:
             INST_NAME("LOCK SUB Eb, Gb");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGB(x2);
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 wback = (nextop&7);    
                 wb2 = (wback>>2);      
@@ -285,14 +285,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x14, 0);
                 B_MARKLOCK(cNE);    // write failed, try again
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x29:
             INST_NAME("LOCK SUB Ed, Gd");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGD;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);
                 emit_sub32(dyn, ninst, ed, gd, x3, x14);
@@ -305,13 +305,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x3, 0);
                 B_MARKLOCK(cNE);
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x38:
             INST_NAME("LOCK XOR Eb, Gb");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 GETGB(x2);
                 wback = (nextop&7);    
@@ -330,14 +330,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x14, 0);
                 B_MARKLOCK(cNE);    // write failed, try again
             }
-            DMB_ISH();
+            SMDMB();
             break;
         case 0x39:
             INST_NAME("LOCK XOR Ed, Gd");
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGD;
-            DMB_ISH();
+            SMDMB();
             if((nextop&0xC0)==0xC0) {
                 ed = xEAX+(nextop&7);
                 emit_xor32(dyn, ninst, ed, gd, x3, x14);
@@ -350,7 +350,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 CMPS_IMM8(x3, 0);
                 B_MARKLOCK(cNE);
             }
-            DMB_ISH();
+            SMDMB();
             break;
         
         case 0x66:
@@ -368,7 +368,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         INST_NAME("LOCK ADD Ed, Ib");
                     }
                     SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         if(opcode==0x81) i32 = F32S; else i32 = F8S;
                         ed = xEAX+(nextop&7);
@@ -389,7 +389,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             B_MARKLOCK(cNE);
                         }
                         if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                             B_NEXT(c__);
                         }
                         if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -403,13 +403,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             B_MARK(cNE);
                             STR_IMM9(x1, wback, 0);    // put the whole value
                         }
-                        DMB_ISH();
+                        SMDMB();
                     }
                     break;
                 case 1: //OR
                     if(opcode==0x81) {INST_NAME("LOCK OR Ed, Id");} else {INST_NAME("LOCK OR Ed, Ib");}
                     SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         if(opcode==0x81) i32 = F32S; else i32 = F8S;
                         ed = xEAX+(nextop&7);
@@ -424,13 +424,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         CMPS_IMM8(x3, 0);
                         B_MARKLOCK(cNE);
                     }
-                    DMB_ISH();
+                    SMDMB();
                     break;
                 case 2: //ADC
                     if(opcode==0x81) {INST_NAME("LOCK ADC Ed, Id");} else {INST_NAME("LOCK ADC Ed, Ib");}
                     READFLAGS(X_CF);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         if(opcode==0x81) i32 = F32S; else i32 = F8S;
                         ed = xEAX+(nextop&7);
@@ -445,13 +445,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         CMPS_IMM8(x3, 0);
                         B_MARKLOCK(cNE);
                     }
-                    DMB_ISH();
+                    SMDMB();
                     break;
                 case 3: //SBB
                     if(opcode==0x81) {INST_NAME("LOCK SBB Ed, Id");} else {INST_NAME("LOCK SBB Ed, Ib");}
                     READFLAGS(X_CF);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         if(opcode==0x81) i32 = F32S; else i32 = F8S;
                         ed = xEAX+(nextop&7);
@@ -466,12 +466,12 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         CMPS_IMM8(x3, 0);
                         B_MARKLOCK(cNE);
                     }
-                    DMB_ISH();
+                    SMDMB();
                     break;
                 case 4: //AND
                     if(opcode==0x81) {INST_NAME("LOCK AND Ed, Id");} else {INST_NAME("LOCK AND Ed, Ib");}
                     SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         if(opcode==0x81) i32 = F32S; else i32 = F8S;
                         ed = xEAX+(nextop&7);
@@ -486,7 +486,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         CMPS_IMM8(x3, 0);
                         B_MARKLOCK(cNE);
                     }
-                    DMB_ISH();
+                    SMDMB();
                     break;
                 case 5: //SUB
                     if(opcode==0x81) {INST_NAME("LOCK SUB Ed, Id");} else {INST_NAME("LOCK SUB Ed, Ib");}
@@ -496,7 +496,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         ed = xEAX+(nextop&7);
                         emit_sub32c(dyn, ninst, ed, i32, x3, x14);
                     } else {
-                        DMB_ISH();
+                        SMDMB();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 1, LOCK_LOCK);
                         if(opcode==0x81) i32 = F32S; else i32 = F8S;
                         if(!fixedaddress) {
@@ -512,7 +512,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             B_MARKLOCK(cNE);
                         }
                         if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                             B_NEXT(c__);
                         }
                         if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -526,7 +526,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             B_MARK(cNE);
                             STR_IMM9(x1, wback, 0);    // put the whole value
                         }
-                        DMB_ISH();
+                        SMDMB();
                     }
                     break;
                 case 6: //XOR
@@ -537,7 +537,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         ed = xEAX+(nextop&7);
                         emit_xor32c(dyn, ninst, ed, i32, x3, x14);
                     } else {
-                        DMB_ISH();
+                        SMDMB();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, LOCK_LOCK);
                         if(opcode==0x81) i32 = F32S; else i32 = F8S;
                         MARKLOCK;
@@ -546,13 +546,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         STREX(x3, x1, wback);
                         CMPS_IMM8(x3, 0);
                         B_MARKLOCK(cNE);
-                        DMB_ISH();
+                        SMDMB();
                     }
                     break;
                 case 7: //CMP
                     if(opcode==0x81) {INST_NAME("(LOCK) CMP Ed, Id");} else {INST_NAME("(LOCK) CMP Ed, Ib");}
                     SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
+                    SMDMB();
                     GETEDH(x1);
                     // No need to LOCK, this is readonly
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -582,7 +582,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         #undef GO
         case 0x86:  // for this two, the lock is already done by the opcode, so just ignoring it
         case 0x87:
-            DMB_ISH();
+            SMDMB();
             --addr;
             break;
         case 0x90:  // This is xchg reg, reg, so let's ignore the LOCK prefix...
@@ -593,7 +593,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0x95:
         case 0x96:
         case 0x97:
-            DMB_ISH();
+            SMDMB();
             --addr;
             break;
     
@@ -603,7 +603,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 case 0xB0:
                     INST_NAME("LOCK CMPXCHG Eb, Gb");
                     SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
+                    SMDMB();
                     nextop = F8;
                     if((nextop&0xC0)==0xC0) {
                         wback = (nextop&7);    
@@ -640,7 +640,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     // AL != Eb
                     BFI(xEAX, ed, 0, 8);
                     MARK3;
-                    DMB_ISH();
+                    SMDMB();
                     // done, do the cmp now
                     UFLAG_IF {emit_cmp8(dyn, ninst, x1, x2, x3, x14);}
                     break;
@@ -649,7 +649,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     nextop = F8;
                     GETGD;
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                         // same as no lock, all in regs...
@@ -679,7 +679,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                                 B_MARK(cNE);
                                 // EAX == Ed
                                 STREX(x14, gd, wback);
-                                DMB_ISH();
+                                SMDMB();
                                 CMPS_IMM8(x14, 0);
                                 B_MARKLOCK(cNE);
                                 B_MARK(c__);
@@ -695,7 +695,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             CMPS_IMM8(x14, 0);
                             B_MARK2(cNE);
                             STR_IMM9(gd, wback, 0);
-                            DMB_ISH();
+                            SMDMB();
                         } else {
                             MARKLOCK;
                             // alignment is good
@@ -706,7 +706,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             STREX(x14, gd, wback);
                             CMPS_IMM8(x14, 0);
                             B_MARKLOCK(cNE);
-                            DMB_ISH();
+                            SMDMB();
                         }
                         // end of EAX == Ed or EAX != Ed
                         MARK;
@@ -721,7 +721,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     SET_DFNONE(x1);
                     nextop = F8;
                     GETGD;
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                         wback = 0;
@@ -747,12 +747,12 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         B_MARKLOCK(cNE);
                     }
                     MARK3;
-                    DMB_ISH();
+                    SMDMB();
                     break;
 
                 case 0xBA:
                     nextop = F8;
-                    DMB_ISH();
+                    SMDMB();
                     switch((nextop>>3)&7) {
                         case 4:
                             INST_NAME("(LOCK) BT Ed, Ib");
@@ -839,7 +839,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         default:
                             DEFAULT;
                     }
-                    DMB_ISH();
+                    SMDMB();
                     break;
                 case 0xBB:
                     INST_NAME("LOCK BTC Ed, Gd");
@@ -847,7 +847,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     SET_DFNONE(x1);
                     nextop = F8;
                     GETGD;
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                         wback = 0;
@@ -870,14 +870,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         CMPS_IMM8(x14, 0);
                         B_MARKLOCK(cNE);
                     }
-                    DMB_ISH();
+                    SMDMB();
                     break;
             
                 case 0xC0:
                     INST_NAME("LOCK XADD Gb, Eb");
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     nextop = F8;
-                    DMB_ISH();
+                    SMDMB();
                     GETGB(x1);
                     if((nextop&0xC0)==0xC0) {
                         wback = (nextop&7); 
@@ -902,14 +902,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     } else {
                         BFI(wback, ed, wb2*8, 8);
                     }
-                    DMB_ISH();
+                    SMDMB();
                     break;
                 case 0xC1:
                     INST_NAME("LOCK XADD Gd, Ed");
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     nextop = F8;
                     GETGD;
-                    DMB_ISH();
+                    SMDMB();
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                         if(gd!=ed) {
@@ -955,7 +955,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         }
                         MOV_REG(gd, x1);
                     }
-                    DMB_ISH();
+                    SMDMB();
                     break;
 
                 case 0xC7:
@@ -964,7 +964,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         case 1:
                             INST_NAME("LOCK CMPXCHG8B Gq, Eq");
                             SETFLAGS(X_ZF, SF_SUBSET);
-                            DMB_ISH();
+                            SMDMB();
                             addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 0, 0, 1, LOCK_LOCK);
                             MARKLOCK;
                             if(!fixedaddress || (fixedaddress && (fixedaddress&7))) {
@@ -1001,7 +1001,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             MOV_REG(xEDX, x3);
                             MOVW(x1, 0);
                             MARK3;
-                            DMB_ISH();
+                            SMDMB();
                             BFI(xFlags, x1, F_ZF, 1);
                             break;
                         default:
@@ -1017,7 +1017,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
         case 0xFF:
             nextop = F8;
-            DMB_ISH();
+            SMDMB();
             switch((nextop>>3)&7)
             {
                 case 0: // INC Ed
@@ -1041,7 +1041,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             B_MARKLOCK(cNE);
                         }
                         if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                             B_NEXT(c__);
                         }
                         if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -1078,7 +1078,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                             B_MARKLOCK(cNE);
                         }
                         if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                             B_NEXT(c__);
                         }
                         if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -1098,7 +1098,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     // default to NO LOCK
                     addr-=2;
             }
-            DMB_ISH();
+            SMDMB();
             break;
        
         default:

src/dynarec/dynarec_arm_f20f.c

@@ -29,6 +29,7 @@
     } else {    \
         parity = getedparity(dyn, ninst, addr, nextop, 3);  \
         a = fpu_get_scratch_double(dyn);            \
+        SMREAD();                                   \
         if(parity) {                                \
             addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL); \
             VLDR_64(a, ed, fixedaddress);           \
@@ -72,6 +73,7 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 VMOVD(v0, d0);
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
@@ -93,6 +95,7 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                 STR_IMM9(x2, ed, fixedaddress+0);
                 STR_IMM9(x3, ed, fixedaddress+4);
+                SMWRITE2();
             }
             break;
         case 0x12:
@@ -104,6 +107,7 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVD(v0, d0);
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
@@ -132,28 +136,50 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             if((nextop&0xC0)==0xC0) {
                 d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 d0 = fpu_get_scratch_double(dyn);
                 VLDR_64(d0, ed, fixedaddress);
             }
+            if(!box86_dynarec_fastround) {
+                VMRS(x14);   // Get FPCSR reg to clear exceptions flags
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
             VCVT_S32_F64(s0, d0);
             VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+                VMSR(x14);  // put back values
+            }
             break;
         case 0x2D:
             INST_NAME("CVTSD2SI Gd, Ex");
-            u8 = sse_setround(dyn, ninst, x1, x2, x14);
+            if(!box86_dynarec_fastround)
+                u8 = sse_setround_reset(dyn, ninst, x1, x2, x14);
+            else
+                u8 = sse_setround(dyn, ninst, x1, x2, x14);
             nextop = F8;
             GETGD;
             s0 = fpu_get_scratch_single(dyn);
             if((nextop&0xC0)==0xC0) {
                 d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 d0 = fpu_get_scratch_double(dyn);
                 VLDR_64(d0, ed, fixedaddress);
             }
             VCVTR_S32_F64(s0, d0);
             VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+            }
             x87_restoreround(dyn, ninst, u8);
             break;
 
@@ -305,6 +331,7 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             if((nextop&0xC0)==0xC0) {
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
@@ -340,6 +367,7 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             if((nextop&0xC0)==0xC0) {
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
@@ -401,6 +429,33 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             VMOV_64(v0, v1);
             break;
 
+        case 0xE6:
+            INST_NAME("CVTPD2DQ Gx, Ex");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            if((nextop&0xC0)==0xC0) {
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                v1 = fpu_get_scratch_quad(dyn);
+                VLD1Q_8(v1, ed);
+            }
+            u8 = sse_setround(dyn, ninst, x1, x2, x14);
+            if(v0<16) {
+                VCVTR_S32_F64(v0*2, v1);
+                VCVTR_S32_F64(v0*2+1, v1+1);
+            } else {
+                d0 = fpu_get_scratch_double(dyn);
+                VCVTR_S32_F64(d0*2, v1);
+                VCVTR_S32_F64(d0*2+1, v1+1);
+                VMOVD(v0, d0);
+            }
+            x87_restoreround(dyn, ninst, u8);
+            VEOR(v0+1, v0+1, v0+1);
+            break;
+
         case 0xF0:
             INST_NAME("LDDQ Gx,Ex");
             nextop = F8;
@@ -410,6 +465,7 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);

src/dynarec/dynarec_arm_f30f.c

@@ -37,6 +37,7 @@
         b = fpu_get_scratch_double(dyn);            \
         a = b*2;                                    \
         parity = getedparity(dyn, ninst, addr, nextop, 2);  \
+        SMREAD();                                   \
         if(parity) {                                \
             addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0, NULL); \
             VLDR_32(a, ed, fixedaddress);           \
@@ -102,6 +103,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 parity = getedparity(dyn, ninst, addr, nextop, 2);
+                SMREAD();
                 if(parity && (v0<16)) {
                     VEORQ(v0, v0, v0);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0, NULL);
@@ -147,6 +149,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0, NULL);
                     STR_IMM9(x2, ed, fixedaddress);
                 }
+                SMWRITE2();
             }
             break;
         case 0x12:
@@ -156,6 +159,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             if((nextop&0xC0)==0xC0) {
                 q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 q1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(q1, ed);
@@ -172,6 +176,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             if((nextop&0xC0)==0xC0) {
                 q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 q1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(q1, ed);
@@ -229,6 +234,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 VMOV_32(s0, d0*2);
             } else {
                 parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
                 if(parity) {
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                     VLDR_32(s0, ed, fixedaddress);
@@ -238,12 +244,27 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     VMOVtoV(s0, x2);
                 }
             }
+            if(!box86_dynarec_fastround) {
+                VMRS(x14);   // Get FPCSR reg to clear exceptions flags
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
             VCVT_S32_F32(s0, s0);
             VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+                VMSR(x14);  // put back values
+            }
             break;
         case 0x2D:
             INST_NAME("CVTSS2SI Gd, Ex");
-            u8 = sse_setround(dyn, ninst, x1, x2, x14);
+            if(!box86_dynarec_fastround)
+                u8 = sse_setround_reset(dyn, ninst, x1, x2, x14);
+            else
+                u8 = sse_setround(dyn, ninst, x1, x2, x14);
             nextop = F8;
             GETGD;
             s0 = fpu_get_scratch_single(dyn);
@@ -258,6 +279,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 VMOV_32(s0, d0*2);
             } else {
                 parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
                 if(parity) {
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                     VLDR_32(s0, ed, fixedaddress);
@@ -269,6 +291,11 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             }
             VCVTR_S32_F32(s0, s0);
             VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+            }
             x87_restoreround(dyn, ninst, u8);
             break;
 
@@ -396,6 +423,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             if((nextop&0xC0)==0xC0) {
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
@@ -500,6 +528,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
@@ -514,14 +543,15 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             INST_NAME("PSHUFHW Gx, Ex, Ib");
             nextop = F8;
             gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
             if((nextop&0xC0)==0xC0) {
                 v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
+                SMREAD();
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
+            v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
             u8 = F8;
             // only high part need to be shuffled. VTBL only handle 8bits value, so the 16bits shuffles need to be changed in 8bits
             u32 = 0;
@@ -554,6 +584,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVD(v0, v1);
             } else {
+                SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 VLD1_64(v0, ed);
@@ -576,6 +607,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 VMOVfrV_D(x2, x3, v0+1);
                 STR_IMM9(x2, ed, fixedaddress+8);
                 STR_IMM9(x3, ed, fixedaddress+12);
+                SMWRITE2();
             }
             break;
 
@@ -640,6 +672,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 VMOV_64(v0, v1);
             } else {
                 parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
                 if(parity) {
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                     VLDR_64(v0, ed, fixedaddress);    // vfpu opcode here
@@ -666,6 +699,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             } else {
                 v0 = fpu_get_scratch_double(dyn);
                 parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
                 if(parity) {
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                     VLDR_64(v0, ed, fixedaddress);    // vfpu opcode here

src/dynarec/dynarec_arm_functions.c

@@ -27,6 +27,8 @@
 #include "dynarec_arm_private.h"
 #include "dynarec_arm_functions.h"
 #include "bridge.h"
+#include "arm_printer.h"
+#include "custommem.h"
 
 void arm_fstp(x86emu_t* emu, void* p)
 {
@@ -109,7 +111,7 @@ void arm_fistp64(x86emu_t* emu, int64_t* ed)
         memcpy(ed, &STll(0).sq, sizeof(int64_t));
     } else {
         int64_t tmp;
-        if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, (double)(int64_t)0x8000000000000000LL) || !isfinite(ST0.d))
+        if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(ST0.d))
             tmp = 0x8000000000000000LL;
         else
             tmp = fpu_round(emu, ST0.d);
@@ -120,7 +122,7 @@ void arm_fistp64(x86emu_t* emu, int64_t* ed)
 int64_t arm_fist64_0(double d)
 {
     int64_t tmp;
-    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
         tmp = 0x8000000000000000LL;
     else
         tmp = nearbyint(d);
@@ -129,7 +131,7 @@ int64_t arm_fist64_0(double d)
 int64_t arm_fist64_1(double d)
 {
     int64_t tmp;
-    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
         tmp = 0x8000000000000000LL;
     else
         tmp = floor(d);
@@ -138,7 +140,7 @@ int64_t arm_fist64_1(double d)
 int64_t arm_fist64_2(double d)
 {
     int64_t tmp;
-    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
         tmp = 0x8000000000000000LL;
     else
         tmp = ceil(d);
@@ -147,7 +149,7 @@ int64_t arm_fist64_2(double d)
 int64_t arm_fist64_3(double d)
 {
     int64_t tmp;
-    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
         tmp = 0x8000000000000000LL;
     else
         tmp = trunc(d);
@@ -170,6 +172,7 @@ void arm_fld(x86emu_t* emu, uint8_t* ed)
 
 void arm_ud(x86emu_t* emu)
 {
+    emu->test.test = 0;
     emit_signal(emu, SIGILL, (void*)R_EIP, 0);
 }
 
@@ -608,6 +611,7 @@ void neoncacheUnwind(neoncache_t* cache)
                 case NEON_CACHE_XMMW:
                     cache->ssecache[cache->neoncache[i].n].reg = i+FPUFIRST;
                     cache->ssecache[cache->neoncache[i].n].write = (cache->neoncache[i].t==NEON_CACHE_XMMW)?1:0;
+                    cache->neoncache[i] = cache->neoncache[i+1];
                     ++cache->fpu_reg;
                     ++i;    // next next
                     break;
@@ -702,13 +706,13 @@ int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int
 #define PK(a)       *(uint8_t*)(addr+a)
 #define PK32(a)     *(uint32_t*)(addr+a)
 
-    if(!addr)
+    if(!addr || !getProtection(addr))
         return 0;
     if(PK(0)==0xff && PK(1)==0x25) {  // absolute jump, maybe the GOT
         uintptr_t a1 = (PK32(2));   // need to add a check to see if the address is from the GOT !
         addr = (uintptr_t)getAlternate(*(void**)a1); 
     }
-    if(addr<0x10000)    // too low, that is suspicious
+    if(addr<0x10000 || !getProtection(addr))    // too low, that is suspicious
         return 0;
     onebridge_t *b = (onebridge_t*)(addr);
     if(b->CC==0xCC && b->S=='S' && b->C=='C' && b->w!=(wrapper_t)0 && b->f!=(uintptr_t)PltResolver) {
@@ -737,24 +741,6 @@ const char* getCacheName(int t, int n)
     return buff;
 }
 
-// is inst clean for a son branch?
-int isInstClean(dynarec_arm_t* dyn, int ninst)
-{
-    // check flags cache
-    if(dyn->insts[ninst].f_entry.dfnone || dyn->insts[ninst].f_entry.pending)
-        return 0;
-    if(dyn->insts[ninst].x86.state_flags)
-        return 0;
-    // check neoncache
-    neoncache_t* n = &dyn->insts[ninst].n;
-    if(n->news || n->stack || n->stack_next)
-        return 0;
-    for(int i=0; i<24; ++i)
-        if(n->neoncache[i].v)
-            return 0;
-    return 1;
-}
-
 static int flagsCacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
     int jmp = dyn->insts[ninst].x86.jmp_insts;
     if(jmp<0)
@@ -795,4 +781,58 @@ int CacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
     if (fpuCacheNeedsTransform(dyn, ninst)) ret|=1;
     if (flagsCacheNeedsTransform(dyn, ninst)) ret|=2;
     return ret;
-}
+}
+
+void inst_name_pass3(dynarec_arm_t* dyn, int ninst, const char* name)
+{
+    if(box86_dynarec_dump) {
+        printf_x86_instruction(my_context->dec, &dyn->insts[ninst].x86, name);
+        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X sm=%d/%d",
+            (box86_dynarec_dump>1)?"\e[32m":"",
+            (void*)(dyn->arm_start+dyn->insts[ninst].address),
+            dyn->insts[ninst].size/4,
+            ninst,
+            dyn->insts[ninst].x86.barrier,
+            dyn->insts[ninst].x86.state_flags,
+            dyn->f.pending,
+            dyn->f.dfnone,
+            dyn->insts[ninst].x86.may_set?"may":"set",
+            dyn->insts[ninst].x86.set_flags,
+            dyn->insts[ninst].x86.gen_flags,
+            dyn->insts[ninst].x86.use_flags,
+            dyn->insts[ninst].x86.need_before,
+            dyn->insts[ninst].x86.need_after,
+            dyn->smread, dyn->smwrite);
+        if(dyn->insts[ninst].pred_sz) {
+            dynarec_log(LOG_NONE, ", pred=");
+            for(int ii=0; ii<dyn->insts[ninst].pred_sz; ++ii)
+                dynarec_log(LOG_NONE, "%s%d", ii?"/":"", dyn->insts[ninst].pred[ii]);
+        }
+        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts>=0)
+            dynarec_log(LOG_NONE, ", jmp=%d", dyn->insts[ninst].x86.jmp_insts);
+        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts==-1)
+            dynarec_log(LOG_NONE, ", jmp=out");
+        for(int ii=0; ii<24; ++ii) {
+            switch(dyn->insts[ninst].n.neoncache[ii].t) {
+                case NEON_CACHE_ST_D: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_ST_F: dynarec_log(LOG_NONE, " S%d:%s", (ii+8)*2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_MM: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_XMMW: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;
+                case NEON_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;
+                case NEON_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_NONE:
+                default:    break;
+            }
+        }
+        if(dyn->n.stack || dyn->insts[ninst].n.stack_next || dyn->insts[ninst].n.x87stack)
+            dynarec_log(LOG_NONE, " X87:%d/%d(+%d/-%d)%d", dyn->n.stack, dyn->insts[ninst].n.stack_next, dyn->insts[ninst].n.stack_push, dyn->insts[ninst].n.stack_pop, dyn->insts[ninst].n.x87stack);
+        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)
+            dynarec_log(LOG_NONE, " %s:%d/%d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);
+        dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");
+    }
+}
+
+void print_opcode(dynarec_arm_t* dyn, int ninst, uint32_t opcode)
+{
+    dynarec_log(LOG_NONE, "\t%08x\t%s\n", opcode, arm_print(opcode));
+}

src/dynarec/dynarec_arm_functions.h

@@ -71,9 +71,6 @@ int CacheNeedsTransform(dynarec_arm_t* dyn, int i1);
 // Undo the changes of a neoncache to get the status before the instruction
 void neoncacheUnwind(neoncache_t* cache);
 
-// is inst clean for a son branch?
-int isInstClean(dynarec_arm_t* dyn, int ninst);
-
 // predecessor access
 int isPred(dynarec_arm_t* dyn, int ninst, int pred);
 int getNominalPred(dynarec_arm_t* dyn, int ninst);
@@ -88,4 +85,7 @@ int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int
 
 const char* getCacheName(int t, int n);
 
+void inst_name_pass3(dynarec_arm_t* dyn, int ninst, const char* name);
+void print_opcode(dynarec_arm_t* dyn, int ninst, uint32_t opcode);
+
 #endif //__DYNAREC_ARM_FUNCTIONS_H__

src/dynarec/dynarec_arm_helper.c

@@ -4,6 +4,7 @@
 #include <pthread.h>
 #include <errno.h>
 #include <assert.h>
+#include <string.h>
 
 #include "debug.h"
 #include "box86context.h"
@@ -185,7 +186,7 @@ uintptr_t geted16(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop,
             case 1: offset = F8S; break;
             case 2: offset = F16S; break;
         }
-        if(offset && (abs(offset)>absmax || (offset&mask))) {
+        if(offset && (abs(offset)<=absmax && !(offset&mask))) {
             *fixaddress = offset;
             offset = 0;
         }
@@ -251,6 +252,7 @@ void jump_to_epilog(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst)
     }
     PASS3(void* epilog = arm_epilog);
     MOV32_(2, (uintptr_t)epilog);
+    SMEND();
     BX(2);
 }
 
@@ -277,6 +279,7 @@ void jump_to_next(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst)
     #ifdef HAVE_TRACE
     MOV_REG(x2, 15);    // move current PC to x2, for tracing
     #endif
+    SMEND();
     BX(x3);
 }
 
@@ -293,6 +296,7 @@ void ret_to_epilog(dynarec_arm_t* dyn, int ninst)
         #ifdef HAVE_TRACE
         MOV_REG(x2, 15);    // move current PC to x2, for tracing
         #endif
+        SMEND();
         BX(x3);
 }
 
@@ -315,6 +319,7 @@ void retn_to_epilog(dynarec_arm_t* dyn, int ninst, int n)
         #ifdef HAVE_TRACE
         MOV_REG(x2, 15);    // move current PC to x2, for tracing
         #endif
+        SMEND();
         BX(x3);
 }
 
@@ -337,6 +342,7 @@ void iret_to_epilog(dynarec_arm_t* dyn, int ninst)
     SET_DFNONE(x1);
     // Ret....
     MOV32_(x2, (uintptr_t)arm_epilog);  // epilog on purpose, CS might have changed!
+    SMEND();
     BX(x2);
 }
 
@@ -478,7 +484,7 @@ void call_rd(dynarec_arm_t* dyn, int ninst, void* fnc, int reg, int s1, uint32_t
         if(s1!=xFlags) mask=1<<xFlags; else mask=1<<x3;
     }
     PUSH(xSP, (1<<xEmu) | mask);
-    fpu_pushcache(dyn, ninst, reg);
+    fpu_pushcache(dyn, ninst, s1);
     if(saveflags) {
         STR_IMM9(xFlags, xEmu, offsetof(x86emu_t, eflags));
     }
@@ -490,7 +496,7 @@ void call_rd(dynarec_arm_t* dyn, int ninst, void* fnc, int reg, int s1, uint32_t
     #ifdef ARM_SOFTFP
     VMOVtoV_64(0, 0, 1);    // load r0:r1 to D0 to simulate hardfo
     #endif
-    fpu_popcache(dyn, ninst, reg);
+    fpu_popcache(dyn, ninst, s1);
     POP(xSP, (1<<xEmu) | mask);
     if(saveflags) {
         LDR_IMM9(xFlags, xEmu, offsetof(x86emu_t, eflags));
@@ -856,10 +862,29 @@ void x87_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int
     MESSAGE(LOG_DUMP, "\t---Purge x87 Cache and Synch Stackcount\n");
 }
 
-#ifdef HAVE_TRACE
 static void x87_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
 {
-    x87_stackcount(dyn, ninst, s1);
+    // Synch top & stack counter
+    int a = dyn->n.x87stack;
+    if(a) {
+    // Add x87stack to emu fpu_stack
+        LDR_IMM9(s2, xEmu, offsetof(x86emu_t, fpu_stack));
+        if(a>0) {
+            ADD_IMM8(s2, s2, a);
+        } else {
+            SUB_IMM8(s2, s2, -a);
+        }
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, fpu_stack));
+        // Sub x87stack to top, with and 7
+        LDR_IMM9(s2, xEmu, offsetof(x86emu_t, top));
+        if(a>0) {
+            SUB_IMM8(s2, s2, a);
+        } else {
+            ADD_IMM8(s2, s2, -a);
+        }
+        AND_IMM8(s2, s2, 7);
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, top));
+    }
     int ret = 0;
     for (int i=0; (i<8) && (!ret); ++i)
         if(dyn->n.x87cache[i] != -1)
@@ -867,13 +892,20 @@ static void x87_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int
     if(!ret)    // nothing to do
         return;
     // Get top
-    LDR_IMM9(s2, xEmu, offsetof(x86emu_t, top));
+    if(!a) {
+        // already there
+        LDR_IMM9(s2, xEmu, offsetof(x86emu_t, top));
+    }
     // loop all cache entries
     for (int i=0; i<8; ++i)
         if(dyn->n.x87cache[i]!=-1) {
-            ADD_IMM8(s3, s2, dyn->n.x87cache[i]);
-            AND_IMM8(s3, s3, 7);    // (emu->top + i)&7
-            ADD_REG_LSL_IMM5(s3, xEmu, s3, 3);    // fpu[(emu->top+i)&7] lsl 3 because fpu are double, so 8 bytes
+            if(dyn->n.x87cache[i]) {
+                ADD_IMM8(s3, s2, dyn->n.x87cache[i]);
+                AND_IMM8(s3, s3, 7);    // (emu->top + i)&7
+                ADD_REG_LSL_IMM5(s3, xEmu, s3, 3);    // fpu[(emu->top+i)&7] lsl 3 because fpu are double, so 8 bytes
+            } else {
+                ADD_REG_LSL_IMM5(s3, xEmu, s2, 3);
+            }
             if(ST_IS_F(dyn->n.x87cache[i])) {
                 VCVT_F64_F32(0, dyn->n.x87reg[i]*2);    // use D0 as scratch...
                 VSTR_64(0, s3, offsetof(x86emu_t, x87));
@@ -882,7 +914,31 @@ static void x87_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int
             }
         }
 }
-#endif
+
+static void x87_unreflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
+{
+    // go back with the top & stack counter
+    int a = dyn->n.x87stack;
+    if(a) {
+    // Add x87stack to emu fpu_stack
+        LDR_IMM9(s1, xEmu, offsetof(x86emu_t, fpu_stack));
+        if(a>0) {
+            SUB_IMM8(s1, s1, a);
+        } else {
+            ADD_IMM8(s1, s1, -a);
+        }
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, fpu_stack));
+        // Sub x87stack to top, with and 7
+        LDR_IMM9(s1, xEmu, offsetof(x86emu_t, top));
+        if(a>0) {
+            ADD_IMM8(s1, s1, a);
+        } else {
+            SUB_IMM8(s1, s1, -a);
+        }
+        AND_IMM8(s1, s1, 7);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, top));
+    }
+}
 
 int x87_get_current_cache(dynarec_arm_t* dyn, int ninst, int st, int t)
 {
@@ -1066,15 +1122,13 @@ void x87_reget_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st)
     MESSAGE(LOG_DUMP, "\t-------x87 Cache for ST%d\n", st);
 }
 
-static int round_map[] = {0, 2, 1, 3};  // map x86 -> arm round flag
-
 // Set rounding according to cw flags, return reg to restore flags
 int x87_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
 {
     LDRH_IMM8(s1, xEmu, offsetof(x86emu_t, cw));    // hopefully cw is not too far for an imm8
-    UBFX(s2, s1, 10, 2);    // extract round...
-    MOV32(s1, round_map);
-    LDR_REG_LSL_IMM5(s2, s1, s2, 2);
+    UBFX(s1, s1, 10, 2);    // extract round...
+    UBFX(s2, s1, 1, 1);     // swap bits 0 and 1
+    BFI(s2, s1, 1, 1);
     VMRS(s1);               // get fpscr
     MOV_REG(s3, s1);
     BFI(s1, s2, 22, 2);     // inject new round
@@ -1082,6 +1136,21 @@ int x87_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
     return s3;
 }
 
+// Set rounding according to cw flags, also enable exception and reset counter, return reg to restore flags
+int x87_setround_reset(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
+{
+    LDRH_IMM8(s1, xEmu, offsetof(x86emu_t, cw));    // hopefully cw is not too far for an imm8
+    UBFX(s1, s1, 10, 2);    // extract round...
+    UBFX(s2, s1, 1, 1);     // swap bits 0 and 1
+    BFI(s2, s1, 1, 1);
+    VMRS(s3);               // get fpscr
+    ORR_IMM8(s1, s3, 0b010, 9); // enable exceptions
+    BIC_IMM8(s1, s1, 0b10011111, 0);    // reset counters
+    BFI(s1, s2, 22, 2);     // inject new round
+    VMSR(s1);               // put new fpscr
+    return s3;
+}
+
 void x87_swapreg(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int b)
 {
     int i1, i2, i3;
@@ -1106,9 +1175,9 @@ void x87_swapreg(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int b)
 int sse_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
 {
     LDR_IMM9(s1, xEmu, offsetof(x86emu_t, mxcsr));
-    UBFX(s2, s1, 13, 2);    // extract round...
-    MOV32(s1, round_map);
-    LDR_REG_LSL_IMM5(s2, s1, s2, 2);
+    UBFX(s1, s1, 13, 2);    // extract round...
+    UBFX(s2, s1, 1, 1);     // swap bits 0 and 1
+    BFI(s2, s1, 1, 1);
     VMRS(s1);               // get fpscr
     MOV_REG(s3, s1);
     BFI(s1, s2, 22, 2);     // inject new round
@@ -1116,6 +1185,21 @@ int sse_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
     return s3;
 }
 
+// Set rounding according to mxcsr flags, also enable exception and reset counter, return reg to restore flags
+int sse_setround_reset(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
+{
+    LDR_IMM9(s1, xEmu, offsetof(x86emu_t, mxcsr));
+    UBFX(s1, s1, 13, 2);    // extract round...
+    UBFX(s2, s1, 1, 1);     // swap bits 0 and 1
+    BFI(s2, s1, 1, 1);
+    VMRS(s3);               // get fpscr
+    ORR_IMM8(s1, s3, 0b010, 9); // enable exceptions
+    BIC_IMM8(s1, s1, 0b10011111, 0);    // reset counters
+    BFI(s1, s2, 22, 2);     // inject new round
+    VMSR(s1);               // put new fpscr
+    return s3;
+}
+
 // Restore round flag
 void x87_restoreround(dynarec_arm_t* dyn, int ninst, int s1)
 {
@@ -1190,7 +1274,6 @@ void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1)
         MESSAGE(LOG_DUMP, "\t------ Purge MMX Cache\n");
     }
 }
-#ifdef HAVE_TRACE
 static void mmx_reflectcache(dynarec_arm_t* dyn, int ninst, int s1)
 {
     int old = -1;
@@ -1208,7 +1291,6 @@ static void mmx_reflectcache(dynarec_arm_t* dyn, int ninst, int s1)
             VST1_32_W(dyn->n.mmxcache[i], s1);
         }
 }
-#endif
 
 
 // SSE / SSE2 helpers
@@ -1288,7 +1370,6 @@ static void sse_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1)
         MESSAGE(LOG_DUMP, "\t------ Purge SSE Cache\n");
     }
 }
-#ifdef HAVE_TRACE
 static void sse_reflectcache(dynarec_arm_t* dyn, int ninst, int s1)
 {
     int old = -1;
@@ -1312,7 +1393,6 @@ static void sse_reflectcache(dynarec_arm_t* dyn, int ninst, int s1)
             VST1Q_32_W(dyn->n.ssecache[i].reg, s1);
         }
 }
-#endif
 
 void fpu_pushcache(dynarec_arm_t* dyn, int ninst, int s1)
 {
@@ -1333,7 +1413,7 @@ void fpu_pushcache(dynarec_arm_t* dyn, int ninst, int s1)
     MOV_REG(s1, xSP);
     for (int i=16; i<32; ++i) {   // should use VSTM?
         if(dyn->n.fpuused[i-8]) {
-            VST1_32_W(i, s1);
+            VST1_64_W(i, s1);
         }
     }
     MESSAGE(LOG_DUMP, "\t------- Push FPU Cache (%d)\n", n);
@@ -1352,7 +1432,7 @@ void fpu_popcache(dynarec_arm_t* dyn, int ninst, int s1)
     MOV_REG(s1, xSP);
     for (int i=16; i<32; ++i) {
         if(dyn->n.fpuused[i-8]) {
-            VLD1_32_W(i, s1);
+            VLD1_64_W(i, s1);
         }
     }
     if(n>=8) {
@@ -1413,8 +1493,10 @@ static void swapCache(dynarec_arm_t* dyn, int ninst, int i, int j, neoncache_t *
         quad =1;
     if(cache->neoncache[j].t==NEON_CACHE_XMMR || cache->neoncache[j].t==NEON_CACHE_XMMW)
         quad =1;
-    if(quad)    // if quad, we need to swap the whole quad!
+    if(quad) {    // if quad, we need to swap the whole quad!
         i&=~1;
+        j&=~1;
+    }
     if(!cache->neoncache[i].v && ((quad && !cache->neoncache[i+1].v) || !quad)) {
         // a mov is enough, no need to swap
         MESSAGE(LOG_DUMP, "\t  - Moving %d <- %d\n", i, j);
@@ -1684,7 +1766,7 @@ static void fpuCacheTransform(dynarec_arm_t* dyn, int ninst, int s1, int s2, int
         stack_cnt = cache_i2.stack;
     }
     neoncache_t cache = dyn->n;
-    // sanitize XMM regs alocation
+    // sanitize XMM regs alocation, cache_i2 has been sanitized in neoncacheUnwind already
     for(int i=0; i<24; i+=2) {
         if(cache.neoncache[i].t == NEON_CACHE_XMMR || cache.neoncache[i].t == NEON_CACHE_XMMW)
             cache.neoncache[i+1] = cache.neoncache[i];
@@ -1792,11 +1874,11 @@ static void flagsCacheTransform(dynarec_arm_t* dyn, int ninst, int s1)
         if(dyn->f.pending!=SF_PENDING) {
             LDR_IMM9(s1, xEmu, offsetof(x86emu_t, df));
             TSTS_REG_LSL_IMM5(s1, s1, 0);
-            j32 = (GETMARK3)-(dyn->arm_size+8);
+            j32 = (GETMARKF2)-(dyn->arm_size+8);
             Bcond(cEQ, j32);
         }
         CALL_(UpdateFlags, -1, 0);
-        MARK3;
+        MARKF2;
     }
 #endif
 }
@@ -1808,16 +1890,17 @@ void CacheTransform(dynarec_arm_t* dyn, int ninst, int cacheupd, int s1, int s2,
         flagsCacheTransform(dyn, ninst, s1);
 }
 
-#ifdef HAVE_TRACE
 void fpu_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
 {
     x87_reflectcache(dyn, ninst, s1, s2, s3);
-    if(trace_emm)
-       mmx_reflectcache(dyn, ninst, s1);
-    if(trace_xmm)
-       sse_reflectcache(dyn, ninst, s1);
+    mmx_reflectcache(dyn, ninst, s1);
+    sse_reflectcache(dyn, ninst, s1);
+}
+
+void fpu_unreflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3)
+{
+    x87_unreflectcache(dyn, ninst, s1, s2, s3);
 }
-#endif
 
 void fpu_reset(dynarec_arm_t* dyn)
 {
@@ -1854,3 +1937,64 @@ void emit_pf(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
     MVN_REG_LSR_REG(s4, s4, s3);
     BFI(xFlags, s4, F_PF, 1);
 }
+
+void fpu_reset_cache(dynarec_arm_t* dyn, int ninst, int reset_n)
+{
+    MESSAGE(LOG_DEBUG, "Reset Caches with %d\n",reset_n);
+    #if STEP > 1
+    // for STEP 2 & 3, just need to refrest with current, and undo the changes (push & swap)
+    dyn->n = dyn->insts[ninst].n;
+    neoncacheUnwind(&dyn->n);
+    #ifdef HAVE_TRACE
+    if(box86_dynarec_dump)
+        if(memcmp(&dyn->n, &dyn->insts[reset_n].n, sizeof(neon_cache_t))) {
+            MESSAGE(LOG_DEBUG, "Warning, difference in neoncache: reset=");
+            for(int i=0; i<24; ++i)
+                if(dyn->insts[reset_n].n.neoncache[i].v)
+                    MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->insts[reset_n].n.neoncache[i].t, dyn->insts[reset_n].n.neoncache[i].n));
+            if(dyn->insts[reset_n].n.combined1 || dyn->insts[reset_n].n.combined2)
+                MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->insts[reset_n].n.swapped?"SWP":"CMB", dyn->insts[reset_n].n.combined1, dyn->insts[reset_n].n.combined2);
+            if(dyn->insts[reset_n].n.stack_push || dyn->insts[reset_n].n.stack_pop)
+                MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->insts[reset_n].n.stack_push, -dyn->insts[reset_n].n.stack_pop);
+            MESSAGE(LOG_DEBUG, " ==> ");
+            for(int i=0; i<24; ++i)
+                if(dyn->insts[ninst].n.neoncache[i].v)
+                    MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->insts[ninst].n.neoncache[i].t, dyn->insts[ninst].n.neoncache[i].n));
+            if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)
+                MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);
+            if(dyn->insts[ninst].n.stack_push || dyn->insts[ninst].n.stack_pop)
+                MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->insts[ninst].n.stack_push, -dyn->insts[ninst].n.stack_pop);
+            MESSAGE(LOG_DEBUG, " -> ");
+            for(int i=0; i<24; ++i)
+                if(dyn->n.neoncache[i].v)
+                    MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->n.neoncache[i].t, dyn->n.neoncache[i].n));
+            if(dyn->n.combined1 || dyn->n.combined2)
+                MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->n.swapped?"SWP":"CMB", dyn->n.combined1, dyn->n.combined2);
+            if(dyn->n.stack_push || dyn->n.stack_pop)
+                MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->n.stack_push, -dyn->n.stack_pop);
+            MESSAGE(LOG_DEBUG, "\n");
+        }
+    #endif //HAVE_TRACE
+    #else
+    dyn->n = dyn->insts[reset_n].n;
+    #endif
+}
+
+// propagate ST stack state, especial stack pop that are deferred
+void fpu_propagate_stack(dynarec_arm_t* dyn, int ninst)
+{
+    if(dyn->n.stack_pop) {
+        for(int j=0; j<24; ++j)
+            if((dyn->n.neoncache[j].t == NEON_CACHE_ST_D || dyn->n.neoncache[j].t == NEON_CACHE_ST_F)) {
+                if(dyn->n.neoncache[j].n<dyn->n.stack_pop)
+                    dyn->n.neoncache[j].v = 0;
+                else
+                    dyn->n.neoncache[j].n-=dyn->n.stack_pop;
+            }
+        dyn->n.stack_pop = 0;
+    }
+    dyn->n.stack = dyn->n.stack_next;
+    dyn->n.news = 0;
+    dyn->n.stack_push = 0;
+    dyn->n.swapped = 0;
+}

src/dynarec/dynarec_arm_helper.h

@@ -27,6 +27,30 @@
 #define PKip(a) *(uint8_t*)(ip+a)
 #define PKa(a)  *(uint8_t*)(a)
 
+// Strong mem emulation helpers
+// Sequence of Read will trigger a DMB on "first" read if strongmem is 2
+// Squence of Write will trigger a DMB on "last" write if strongmem is 1
+// Opcode will read
+#define SMREAD()    if(!dyn->smread && box86_dynarec_strongmem>1) {SMDMB();}
+// Opcode will read with option forced lock
+#define SMREADLOCK(lock)    if(lock || (!dyn->smread && box86_dynarec_strongmem>1)) {SMDMB();}
+// Opcode migh read (depend on nextop)
+#define SMMIGHTREAD()   if(!MODREG) {SMREAD();}
+// Opcode has wrote
+#define SMWRITE()   dyn->smwrite=1
+// Opcode has wrote (strongmem>1 only)
+#define SMWRITE2()   if(box86_dynarec_strongmem>1) dyn->smwrite=1
+// Opcode has wrote with option forced lock
+#define SMWRITELOCK(lock)   if(lock) {SMDMB();} else dyn->smwrite=1
+// Opcode migh have wrote (depend on nextop)
+#define SMMIGHTWRITE()   if(!MODREG) {SMWRITE();}
+// Start of sequence
+#define SMSTART()   SMEND()
+// End of sequence
+#define SMEND()     if(dyn->smwrite && box86_dynarec_strongmem) {DMB_ISH();} dyn->smwrite=0; dyn->smread=0;
+// Force a Data memory barrier (for LOCK: prefix)
+#define SMDMB()     DMB_ISH(); dyn->smwrite=0; dyn->smread=1
+
 //LOCK_* define
 #define LOCK_LOCK   (int*)1
 
@@ -37,6 +61,7 @@
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0, NULL); \
                     LDR_IMM9(x1, wback, fixedaddress); \
                     ed = x1;                \
@@ -46,6 +71,7 @@
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, (hint==x2)?x1:x2, &fixedaddress, 4095, 0, 0, NULL); \
                     LDR_IMM9(hint, wback, fixedaddress); \
                     ed = hint;              \
@@ -56,19 +82,21 @@
                     MOV_REG(ret, ed);       \
                     wback = 0;              \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, 4095, 0, 0, NULL); \
                     ed = ret;               \
                     LDR_IMM9(ed, wback, fixedaddress); \
                 }
 // Write back ed in wback (if wback not 0)
-#define WBACK       if(wback) {STR_IMM9(ed, wback, fixedaddress);}
-// Send back wb to either ed or wback
-#define SBACK(wb)   if(wback) {STR_IMM9(wb, wback, fixedaddress);} else {MOV_REG(ed, wb);}
+#define WBACK       if(wback) {STR_IMM9(ed, wback, fixedaddress); SMWRITE();}
+// Write back ed in wback (if wback not 0) (SMWRITE2 version)
+#define WBACK2      if(wback) {STR_IMM9(ed, wback, fixedaddress); SMWRITE2();}
 //GETEDO can use r1 for ed, and r2 for wback. wback is 0 if ed is xEAX..xEDI
 #define GETEDO(O)   if((nextop&0xC0)==0xC0) {   \
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, NULL); \
                     LDR_REG_LSL_IMM5(x1, wback, O, 0);  \
                     ed = x1;                 \
@@ -78,13 +106,14 @@
                     ed = xEAX+(nextop&7);               \
                     wback = 0;                          \
                 } else {                                \
+                    SMREAD();                           \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, NULL); \
                     ADD_REG_LSL_IMM5(x2, wback, O, 0);  \
                     if(wback != x2) wback = x2;         \
                     LDR_IMM9(x1, wback, 0);             \
                     ed = x1;                            \
                 }
-#define WBACKO(O)   if(wback) {STR_REG_LSL_IMM5(ed, wback, O, 0);}
+#define WBACKO(O)   if(wback) {STR_REG_LSL_IMM5(ed, wback, O, 0); SMWRITE();}
 //FAKEELike GETED, but doesn't get anything
 #define FAKEED  if((nextop&0xC0)!=0xC0) {   \
                     addr = fakeed(dyn, addr, ninst, nextop); \
@@ -98,6 +127,7 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, w, &fixedaddress, 255, 0, 0, NULL); \
                     LDRH_IMM8(i, wback, fixedaddress); \
                     ed = i;                 \
@@ -110,6 +140,7 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRH_IMM8(i, wback, fixedaddress); \
                     ed = i;                 \
@@ -122,15 +153,16 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRSH_IMM8(i, wback, fixedaddress);\
                     ed = i;                 \
                     wb1 = 1;                \
                 }
 // Write ed back to original register / memory
-#define EWBACK   if(wb1) {STRH_IMM8(ed, wback, fixedaddress);} else {BFI(wback, ed, 0, 16);}
+#define EWBACK   if(wb1) {STRH_IMM8(ed, wback, fixedaddress); SMWRITE();} else {BFI(wback, ed, 0, 16);}
 // Write w back to original register / memory
-#define EWBACKW(w)   if(wb1) {STRH_IMM8(w, wback, fixedaddress);} else {BFI(wback, w, 0, 16);}
+#define EWBACKW(w)   if(wb1) {STRH_IMM8(w, wback, fixedaddress); SMWRITE();} else {BFI(wback, w, 0, 16);}
 // Write back gd in correct register
 #define GWBACK       BFI((xEAX+((nextop&0x38)>>3)), gd, 0, 16);
 //GETEB will use i for ed, and can use r3 for wback.
@@ -142,6 +174,7 @@
                     wb1 = 0;                \
                     ed = i;                 \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL); \
                     LDRB_IMM9(i, wback, fixedaddress); \
                     wb1 = 1;                \
@@ -156,6 +189,7 @@
                     wb1 = 0;                \
                     ed = i;                 \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0, NULL); \
                     ADD_REG_LSL_IMM5(x3, wback, i, 0);  \
                     if(wback!=x3) wback = x3;           \
@@ -172,13 +206,16 @@
                     wb1 = 0;                \
                     ed = i;                 \
                 } else {                    \
+                    SMREAD();               \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRSB_IMM8(i, wback, fixedaddress);\
                     wb1 = 1;                \
                     ed = i;                 \
                 }
 // Write eb (ed) back to original register / memory
-#define EBBACK   if(wb1) {STRB_IMM9(ed, wback, fixedaddress);} else {BFI(wback, ed, wb2*8, 8);}
+#define EBBACK   if(wb1) {STRB_IMM9(ed, wback, fixedaddress); SMWRITE();} else {BFI(wback, ed, wb2*8, 8);}
+// Write eb (ed) back to original register / memory (SMWRITE2 version)
+#define EBBACK2   if(wb1) {STRB_IMM9(ed, wback, fixedaddress); SMWRITE2();} else {BFI(wback, ed, wb2*8, 8);}
 //GETGB will use i for gd
 #define GETGB(i)    gd = (nextop&0x38)>>3;  \
                     gb2 = ((gd&4)>>2);      \
@@ -223,17 +260,19 @@
 // CALL_1RD will use S as scratch. Return value in D0, 1 ARG in R
 #define CALL_1RD(F, R, S, M) call_rd(dyn, ninst, F, R, S, M, 0);
 
-#define MARK    if(dyn->insts) {dyn->insts[ninst].mark = (uintptr_t)dyn->arm_size;}
-#define GETMARK ((dyn->insts)?dyn->insts[ninst].mark:(dyn->arm_size+4))
-#define MARK2   if(dyn->insts) {dyn->insts[ninst].mark2 = (uintptr_t)dyn->arm_size;}
-#define GETMARK2 ((dyn->insts)?dyn->insts[ninst].mark2:(dyn->arm_size+4))
-#define MARK3   if(dyn->insts) {dyn->insts[ninst].mark3 = (uintptr_t)dyn->arm_size;}
-#define GETMARK3 ((dyn->insts)?dyn->insts[ninst].mark3:(dyn->arm_size+4))
-#define MARKF   if(dyn->insts) {dyn->insts[ninst].markf = (uintptr_t)dyn->arm_size;}
-#define GETMARKF ((dyn->insts)?dyn->insts[ninst].markf:(dyn->arm_size+4))
-#define MARKSEG if(dyn->insts) {dyn->insts[ninst].markseg = (uintptr_t)dyn->arm_size;}
-#define GETMARKSEG ((dyn->insts)?dyn->insts[ninst].markseg:(dyn->arm_size+4))
-#define MARKLOCK if(dyn->insts) {dyn->insts[ninst].marklock = (uintptr_t)dyn->arm_size;}
+#define MARK        if(dyn->insts) {dyn->insts[ninst].mark = (uintptr_t)dyn->arm_size;}
+#define GETMARK     ((dyn->insts)?dyn->insts[ninst].mark:(dyn->arm_size+4))
+#define MARK2       if(dyn->insts) {dyn->insts[ninst].mark2 = (uintptr_t)dyn->arm_size;}
+#define GETMARK2    ((dyn->insts)?dyn->insts[ninst].mark2:(dyn->arm_size+4))
+#define MARK3       if(dyn->insts) {dyn->insts[ninst].mark3 = (uintptr_t)dyn->arm_size;}
+#define GETMARK3    ((dyn->insts)?dyn->insts[ninst].mark3:(dyn->arm_size+4))
+#define MARKF       if(dyn->insts) {dyn->insts[ninst].markf = (uintptr_t)dyn->arm_size;}
+#define GETMARKF    ((dyn->insts)?dyn->insts[ninst].markf:(dyn->arm_size+4))
+#define MARKF2      if(dyn->insts) {dyn->insts[ninst].markf2 = (uintptr_t)dyn->arm_size;}
+#define GETMARKF2   ((dyn->insts)?dyn->insts[ninst].markf2:(dyn->arm_size+4))
+#define MARKSEG     if(dyn->insts) {dyn->insts[ninst].markseg = (uintptr_t)dyn->arm_size;}
+#define GETMARKSEG  ((dyn->insts)?dyn->insts[ninst].markseg:(dyn->arm_size+4))
+#define MARKLOCK    if(dyn->insts) {dyn->insts[ninst].marklock = (uintptr_t)dyn->arm_size;}
 #define GETMARKLOCK ((dyn->insts)?dyn->insts[ninst].marklock:(dyn->arm_size+4))
 
 // Branch to MARK if cond (use j32)
@@ -306,7 +345,7 @@
 
 #ifndef READFLAGS
 #define READFLAGS(A) \
-    if((dyn->f.pending!=SF_SET)                         \
+    if(((A)!=X_PEND && dyn->f.pending!=SF_SET)          \
     && (dyn->f.pending!=SF_SET_PENDING)) {              \
         if(dyn->f.pending!=SF_PENDING) {                \
             LDR_IMM9(x3, xEmu, offsetof(x86emu_t, df)); \
@@ -497,9 +536,11 @@ void* arm_next(x86emu_t* emu, uintptr_t addr);
 #define x87_reget_st    STEPNAME(x87_reget_st)
 #define x87_stackcount  STEPNAME(x87_stackcount)
 #define x87_setround    STEPNAME(x87_setround)
+#define x87_setround_reset    STEPNAME(x87_setround_reset)
 #define x87_restoreround STEPNAME(x87_restoreround)
 #define x87_swapreg     STEPNAME(x87_swapreg)
 #define sse_setround    STEPNAME(sse_setround)
+#define sse_setround_reset    STEPNAME(sse_setround_reset)
 #define mmx_get_reg     STEPNAME(mmx_get_reg)
 #define mmx_get_reg_empty STEPNAME(mmx_get_reg_empty)
 #define sse_get_reg     STEPNAME(sse_get_reg)
@@ -508,12 +549,13 @@ void* arm_next(x86emu_t* emu, uintptr_t addr);
 #define fpu_pushcache   STEPNAME(fpu_pushcache)
 #define fpu_popcache    STEPNAME(fpu_popcache)
 #define fpu_reset       STEPNAME(fpu_reset)
+#define fpu_reset_cache STEPNAME(fpu_reset_cache)
+#define fpu_propagate_stack STEPNAME(fpu_propagate_stack)
 #define fpu_purgecache  STEPNAME(fpu_purgecache)
 #define x87_purgecache  STEPNAME(x87_purgecache)
 #define mmx_purgecache  STEPNAME(mmx_purgecache)
-#ifdef HAVE_TRACE
 #define fpu_reflectcache STEPNAME(fpu_reflectcache)
-#endif
+#define fpu_unreflectcache STEPNAME(fpu_unreflectcache)
 
 // get the single reg that from the double "reg" (so Dx[idx])
 #define fpu_get_single_reg      STEPNAME(fpu_get_single_reg)
@@ -644,12 +686,16 @@ void x87_forget(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 void x87_reget_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 // Set rounding according to cw flags, return reg to restore flags
 int x87_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+// Set rounding according to cw flags, return reg to restore flags, also enable exceptions and reset counters
+int x87_setround_reset(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 // Restore round flag
 void x87_restoreround(dynarec_arm_t* dyn, int ninst, int s1);
 // swap 2 x87 regs
 void x87_swapreg(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int b);
 // Set rounding according to mxcsr flags, return reg to restore flags
 int sse_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+// Set rounding according to mxcsr flags, return reg to restore flags, also enable exceptions and reset counters
+int sse_setround_reset(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 
 void CacheTransform(dynarec_arm_t* dyn, int ninst, int cacheupd, int s1, int s2, int s3);
 
@@ -698,13 +744,16 @@ int sse_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int a);
 // common coproc helpers
 // reset the cache
 void fpu_reset(dynarec_arm_t* dyn);
+// reset the cache with n
+void fpu_reset_cache(dynarec_arm_t* dyn, int ninst, int reset_n);
+// propagate stack state
+void fpu_propagate_stack(dynarec_arm_t* dyn, int ninst);
 // purge the FPU cache (needs 3 scratch registers) next=1 if for a conditionnal branch jumping out of block (no tracking updated)
 void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
 void x87_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
 void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1);
-#ifdef HAVE_TRACE
 void fpu_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
-#endif
+void fpu_unreflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 void fpu_pushcache(dynarec_arm_t* dyn, int ninst, int s1);
 void fpu_popcache(dynarec_arm_t* dyn, int ninst, int s1);
 
@@ -751,4 +800,21 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
 #define MAYUSE(A)   
 #endif
 
+#define NOTEST(s1)                                          \
+    if(box86_dynarec_test) {                                \
+        MOVW(s1, 0);                                        \
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, test.test));  \
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, test.clean)); \
+    }
+#define SKIPTEST(s1)                                        \
+    if(box86_dynarec_test) {                                \
+        MOVW(s1, 0);                                        \
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, test.clean)); \
+    }
+#define GOTEST(s1, s2)                                      \
+    if(box86_dynarec_test) {                                \
+        MOVW(s2, 1);                                        \
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, test.test));  \
+    }
+
 #endif //__DYNAREC_ARM_HELPER_H__

src/dynarec/dynarec_arm_jmpnext.c

@@ -0,0 +1,12 @@
+#include <stdint.h>
+#include <stdlib.h>
+
+#include "arm_emitter.h"
+
+#define EMIT(A) *block = (A); ++block
+void CreateJmpNext(void* addr, void* next)
+{
+    uint32_t* block = (uint32_t*)addr;
+    LDR_literal(x2, ((intptr_t)next - (intptr_t)addr)-8);
+    BX(x2);
+}

src/dynarec/dynarec_arm_pass.c

@@ -21,6 +21,7 @@
 #include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"
 #include "custommem.h"
+#include "elfloader.h"
 
 #ifndef STEP
 #error No STEP defined
@@ -35,13 +36,16 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
     uintptr_t init_addr = addr;
     MAYUSE(init_addr);
     int need_epilog = 1;
-    dyn->sons_size = 0;
     // Clean up (because there are multiple passes)
     dyn->f.pending = 0;
     dyn->f.dfnone = 0;
+    dyn->forward = 0;
+    dyn->forward_to = 0;
+    dyn->forward_size = 0;
+    dyn->forward_ninst = 0;
     fpu_reset(dyn);
     int reset_n = -1;
-    // ok, go now
+    int stopblock = 2+(FindElfAddress(my_context, addr)?0:1); // if block is in elf_memory, it can be extended with bligblocks==2, else it needs 3    // ok, go now
     INIT;
     while(ok) {
         ip = addr;
@@ -52,44 +56,7 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
                 dyn->f.pending = 0;
                 fpu_reset(dyn);
             } else {
-                MESSAGE(LOG_DEBUG, "Reset Caches with %d\n",reset_n);
-                #if STEP > 1
-                // for STEP 2 & 3, just need to refrest with current, and undo the changes (push & swap)
-                dyn->n = dyn->insts[ninst].n;
-                neoncacheUnwind(&dyn->n);
-                #ifdef HAVE_TRACE
-                if(box86_dynarec_dump)
-                    if(memcmp(&dyn->n, &dyn->insts[reset_n].n, sizeof(neon_cache_t))) {
-                        MESSAGE(LOG_DEBUG, "Warning, difference in neoncache: reset=");
-                        for(int i=0; i<24; ++i)
-                            if(dyn->insts[reset_n].n.neoncache[i].v)
-                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->insts[reset_n].n.neoncache[i].t, dyn->insts[reset_n].n.neoncache[i].n));
-                        if(dyn->insts[reset_n].n.combined1 || dyn->insts[reset_n].n.combined2)
-                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->insts[reset_n].n.swapped?"SWP":"CMB", dyn->insts[reset_n].n.combined1, dyn->insts[reset_n].n.combined2);
-                        if(dyn->insts[reset_n].n.stack_push || dyn->insts[reset_n].n.stack_pop)
-                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->insts[reset_n].n.stack_push, -dyn->insts[reset_n].n.stack_pop);
-                        MESSAGE(LOG_DEBUG, " ==> ");
-                        for(int i=0; i<24; ++i)
-                            if(dyn->insts[ninst].n.neoncache[i].v)
-                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->insts[ninst].n.neoncache[i].t, dyn->insts[ninst].n.neoncache[i].n));
-                        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)
-                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);
-                        if(dyn->insts[ninst].n.stack_push || dyn->insts[ninst].n.stack_pop)
-                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->insts[ninst].n.stack_push, -dyn->insts[ninst].n.stack_pop);
-                        MESSAGE(LOG_DEBUG, " -> ");
-                        for(int i=0; i<24; ++i)
-                            if(dyn->n.neoncache[i].v)
-                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->n.neoncache[i].t, dyn->n.neoncache[i].n));
-                        if(dyn->n.combined1 || dyn->n.combined2)
-                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->n.swapped?"SWP":"CMB", dyn->n.combined1, dyn->n.combined2);
-                        if(dyn->n.stack_push || dyn->n.stack_pop)
-                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->n.stack_push, -dyn->n.stack_pop);
-                        MESSAGE(LOG_DEBUG, "\n");
-                    }
-                #endif //HAVE_TRACE
-                #else
-                dyn->n = dyn->insts[reset_n].n;
-                #endif
+                fpu_reset_cache(dyn, ninst, reset_n);
                 dyn->f = dyn->insts[reset_n].f_exit;
                 if(dyn->insts[ninst].x86.barrier&BARRIER_FLOAT) {
                     MESSAGE(LOG_DEBUG, "Apply Barrier Float\n");
@@ -103,28 +70,29 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
             }
             reset_n = -1;
         }
-        // propagate ST stack state, especial stack pop that are defered
-        if(dyn->n.stack_pop) {
-            for(int j=0; j<24; ++j)
-                if((dyn->n.neoncache[j].t == NEON_CACHE_ST_D || dyn->n.neoncache[j].t == NEON_CACHE_ST_F)) {
-                    if(dyn->n.neoncache[j].n<dyn->n.stack_pop)
-                        dyn->n.neoncache[j].v = 0;
-                    else
-                        dyn->n.neoncache[j].n-=dyn->n.stack_pop;
-                }
-            dyn->n.stack_pop = 0;
-        }
-        dyn->n.stack = dyn->n.stack_next;
-        dyn->n.news = 0;
-        dyn->n.stack_push = 0;
-        dyn->n.swapped = 0;
+        fpu_propagate_stack(dyn, ninst);
         NEW_INST;
+        if(!ninst) {
+            GOTEST(x1, x2);
+        }
+        if(dyn->insts[ninst].pred_sz>1) {SMSTART();}
         fpu_reset_scratch(dyn);
         if((dyn->insts[ninst].x86.need_before&~X_PEND) && !dyn->insts[ninst].pred_sz) {
             READFLAGS(dyn->insts[ninst].x86.need_before&~X_PEND);
         }
+        if(box86_dynarec_test)  {
+            MESSAGE(LOG_DUMP, "TEST INIT  ----\n");
+            fpu_reflectcache(dyn, ninst, x1, x2, x3);
+            MOV32(x1, ip);
+            STM(xEmu, (1<<xEAX)|(1<<xEBX)|(1<<xECX)|(1<<xEDX)|(1<<xESI)|(1<<xEDI)|(1<<xESP)|(1<<xEBP));
+            STR_IMM9(x1, xEmu, offsetof(x86emu_t, ip));
+            MOVW(x2, 1);
+            CALL(x86test_init, -1, 0);
+            fpu_unreflectcache(dyn, ninst, x1, x2, x3);
+            MESSAGE(LOG_DUMP, "----------\n");
+        }
 #ifdef HAVE_TRACE
-        if(my_context->dec && box86_dynarec_trace) {
+        else if(my_context->dec && box86_dynarec_trace) {
         if((trace_end == 0) 
             || ((ip >= trace_start) && (ip < trace_end)))  {
                 MESSAGE(LOG_DUMP, "TRACE ----\n");
@@ -134,6 +102,7 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
                 STR_IMM9(x1, xEmu, offsetof(x86emu_t, ip));
                 MOVW(x2, 1);
                 CALL(PrintTrace, -1, 0);
+                fpu_unreflectcache(dyn, ninst, x1, x2, x3);
                 MESSAGE(LOG_DUMP, "----------\n");
             }
         }
@@ -156,13 +125,26 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
                 dyn->f.dfnone = 0;
             }
         }
+        #ifndef PROT_READ
+        #define PROT_READ 1
+        #endif
         #if STEP != 0
         if(!ok && !need_epilog && (addr < (dyn->start+dyn->isize))) {
             ok = 1;
             // we use the 1st predecessor here
             int ii = ninst+1;
-            while(ii<dyn->size && !dyn->insts[ii].pred_sz) 
-                ++ii;
+            if(ii<dyn->size && !dyn->insts[ii].pred_sz) {
+                while(ii<dyn->size && (!dyn->insts[ii].pred_sz || (dyn->insts[ii].pred_sz==1 && dyn->insts[ii].pred[0]==ii-1))) {
+                    // may need to skip opcodes to advance
+                    ++ninst;
+                    NEW_INST;
+                    MESSAGE(LOG_DEBUG, "Skipping unused opcode\n");
+                    INST_NAME("Skipped opcode");
+                    INST_EPILOG;
+                    addr += dyn->insts[ii].x86.size;
+                    ++ii;
+                }
+            }
             if((dyn->insts[ii].x86.barrier&BARRIER_FULL)==BARRIER_FULL)
                 reset_n = -2;    // hack to say Barrier!
             else {
@@ -174,15 +156,33 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
             }
         }
         #else
-        #ifndef PROT_READ
-        #define PROT_READ 1
-        #endif
-        if(!ok && !need_epilog && box86_dynarec_bigblock && (getProtection(addr+3)&PROT_READ))
+        if(dyn->forward) {
+            if(dyn->forward_to == addr && !need_epilog) {
+                // we made it!
+                if(box86_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);
+                dyn->forward = 0;
+                dyn->forward_to = 0;
+                dyn->forward_size = 0;
+                dyn->forward_ninst = 0;
+                ok = 1; // in case it was 0
+            } else if ((dyn->forward_to < addr) || !ok) {
+                // something when wrong! rollback
+                if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Could not forward extend block for %d bytes %p -> %p\n", dyn->forward_to-dyn->forward, (void*)dyn->forward, (void*)dyn->forward_to);
+                ok = 0;
+                dyn->size = dyn->forward_size;
+                ninst = dyn->forward_ninst;
+                addr = dyn->forward;
+                dyn->forward = 0;
+                dyn->forward_to = 0;
+                dyn->forward_size = 0;
+                dyn->forward_ninst = 0;
+            }
+            // else just continue
+        } else if(!ok && !need_epilog && box86_dynarec_bigblock && (getProtection(addr+3)&~PROT_READ))
             if(*(uint32_t*)addr!=0) {   // check if need to continue (but is next 4 bytes are 0, stop)
                 uintptr_t next = get_closest_next(dyn, addr);
                 if(next && (
                     (((next-addr)<15) && is_nops(dyn, addr, next-addr)) 
-                    && (((box86_dynarec_bigblock<2) && isJumpTableDefault((void*)next)) || (box86_dynarec_bigblock>1))
                     /*||(((next-addr)<30) && is_instructions(dyn, addr, next-addr))*/ ))
                 {
                     ok = 1;
@@ -194,23 +194,56 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
                             ii=ninst;
                         }
                     if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Extend block %p, %p -> %p (ninst=%d, jump from %d)\n", dyn, (void*)addr, (void*)next, ninst, reset_n);
-                } else if(next && (next-addr)<30) {
-                    if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Cannot extend block %p -> %p (%02X %02X %02X %02X %02X %02X %02X %02x)\n", (void*)addr, (void*)next, PK(0), PK(1), PK(2), PK(3), PK(4), PK(5), PK(6), PK(7));
+                } else if(next && (next-addr)<box86_dynarec_forward && (getProtection(next)&PROT_READ)/*box86_dynarec_bigblock>=stopblock*/) {
+                    if(!((box86_dynarec_bigblock<stopblock) && !isJumpTableDefault((void*)next))) {
+                        dyn->forward = addr;
+                        dyn->forward_to = next;
+                        dyn->forward_size = dyn->size;
+                        dyn->forward_ninst = ninst;
+                        reset_n = -2;
+                        ok = 1;
+                    }
                 }
             }
         #endif
-        if(ok<0)  {ok = 0; need_epilog=1;}
+        if(ok<0)  {
+            ok = 0; need_epilog=1; 
+            #if STEP == 0
+            if(ninst) {
+                --ninst;
+                if(!dyn->insts[ninst].x86.barrier) {
+                    BARRIER(BARRIER_FLOAT);
+                }
+                dyn->insts[ninst].x86.need_after |= X_PEND;
+                ++ninst;
+            }
+            #endif
+        }
         ++ninst;
         #if STEP == 0
-        if((ok && !isJumpTableDefault((void*)addr) && (box86_dynarec_bigblock<2)) ||
-            (ok && dyn->size >= 2000))
+        if(ok && (((box86_dynarec_bigblock<stopblock) && !isJumpTableDefault((void*)addr)) 
+            || (addr>=box86_nodynarec_start && addr<box86_nodynarec_end)))
         #else
         if(ok && (ninst==dyn->size))
         #endif
         {
+            #if STEP == 0
+            if(dyn->forward) {
+                // stopping too soon
+                dyn->size = dyn->forward_size;
+                ninst = dyn->forward_ninst+1;
+                addr = dyn->forward;
+                dyn->forward = 0;
+                dyn->forward_to = 0;
+                dyn->forward_size = 0;
+                dyn->forward_ninst = 0;
+            }
+            #endif
             int j32;
             MAYUSE(j32);
-            MESSAGE(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size); 
+            MESSAGE(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size);
+            if(!box86_dynarec_dump && addr>=box86_nodynarec_start && addr<box86_nodynarec_end)
+                dynarec_log(LOG_INFO, "Stopping block in no-dynarec zone\n");
             --ninst;
             if(!dyn->insts[ninst].x86.barrier) {
                 BARRIER(BARRIER_FLOAT);
@@ -219,16 +252,18 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
             dyn->insts[ninst].x86.need_after |= X_PEND;
             #endif
             ++ninst;
+            NOTEST(x3);
             fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
             jump_to_next(dyn, addr, 0, ninst);
             ok=0; need_epilog=0;
         }
     }
     if(need_epilog) {
+        NOTEST(x3);
         fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
         jump_to_epilog(dyn, ip, 0, ninst);  // no linker here, it's an unknow instruction
     }
     FINI;
-    MESSAGE(LOG_DUMP, "---- END OF BLOCK ---- (%d, %d sons)\n", dyn->size, dyn->sons_size);
+    MESSAGE(LOG_DUMP, "---- END OF BLOCK ---- (%d)\n", dyn->size);
     return addr;
 }

src/dynarec/dynarec_arm_pass0.h

@@ -4,6 +4,7 @@
     dyn->isize = addr-sav_addr;\
     dyn->insts[ninst].x86.addr = addr;\
     if(ninst) dyn->insts[ninst-1].x86.size = dyn->insts[ninst].x86.addr - dyn->insts[ninst-1].x86.addr
+
 #define MESSAGE(A, ...)  
 #define MAYSETFLAGS()   dyn->insts[ninst].x86.may_set = 1
 #define READFLAGS(A)    \
@@ -15,13 +16,13 @@
         dyn->f.pending=(B)&SF_SET_PENDING;      \
         dyn->f.dfnone=((B)&SF_SET)?1:0;
 #define EMIT(A)     
-#define JUMP(A, C)      if((A)>addr) add_next(dyn, (uintptr_t)(A)); dyn->insts[ninst].x86.jmp = A; dyn->insts[ninst].x86.jmp_cond = C
+#define JUMP(A, C)      add_next(dyn, (uintptr_t)A); dyn->insts[ninst].x86.jmp = A; dyn->insts[ninst].x86.jmp_cond = C
 #define BARRIER(A)      if(A!=BARRIER_MAYBE) {fpu_purgecache(dyn, ninst, 0, x1, x2, x3); dyn->insts[ninst].x86.barrier = A;} else dyn->insts[ninst].barrier_maybe = 1
 #define BARRIER_NEXT(A) dyn->insts[ninst+1].x86.barrier = A
 #define NEW_INST \
         ++dyn->size;                            \
         if(dyn->size+3>=dyn->cap) {             \
-                dyn->insts = (instruction_arm_t*)box_realloc(dyn->insts, sizeof(instruction_arm_t)*dyn->cap*2); \
+                dyn->insts = (instruction_arm_t*)customRealloc(dyn->insts, sizeof(instruction_arm_t)*dyn->cap*2);\
                 memset(&dyn->insts[dyn->cap], 0, sizeof(instruction_arm_t)*dyn->cap);   \
                 dyn->cap *= 2;                  \
         }                                       \
@@ -34,7 +35,7 @@
 #define INST_EPILOG                             \
         dyn->insts[ninst].f_exit = dyn->f;      \
         dyn->insts[ninst].n = dyn->n;           \
-        dyn->insts[ninst].x86.has_next = ok;
+        dyn->insts[ninst].x86.has_next = (ok>0)?1:0;
 #define INST_NAME(name) 
 #define DEFAULT                         \
         --dyn->size;                    \

src/dynarec/dynarec_arm_pass1.h

@@ -2,7 +2,6 @@
 #define FINI
 #define MESSAGE(A, ...)  
 #define EMIT(A)     
-
 #define NEW_INST                                \
         dyn->insts[ninst].f_entry = dyn->f;     \
         dyn->n.combined1 = dyn->n.combined2 = 0;\

src/dynarec/dynarec_arm_pass2.h

@@ -1,13 +1,16 @@
 #define INIT        dyn->arm_size = 0
-#define FINI        if(ninst) {dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);}
+#define FINI                                                                                            \
+        if(ninst) {                                                                                     \
+                dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);     \
+                dyn->insts_size += 1+((dyn->insts[ninst].x86.size>dyn->insts[ninst].size)?dyn->insts[ninst].x86.size:dyn->insts[ninst].size)/15; \
+        }
 
 #define MESSAGE(A, ...)  
-#define EMIT(A)     dyn->insts[ninst].size+=4; dyn->arm_size+=4
+#define EMIT(A)     do{dyn->insts[ninst].size+=4; dyn->arm_size+=4;}while(0)
 #define NEW_INST                                                                                        \
         if(ninst) {                                                                                     \
                 dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);     \
-                if(ninst && isInstClean(dyn, ninst))                                                    \
-                        ++dyn->sons_size;                                                               \
+                dyn->insts_size += 1+((dyn->insts[ninst-1].x86.size>dyn->insts[ninst-1].size)?dyn->insts[ninst-1].x86.size:dyn->insts[ninst-1].size)/15; \
         }
 #define INST_EPILOG dyn->insts[ninst].epilog = dyn->arm_size; 
 #define INST_NAME(name) 

src/dynarec/dynarec_arm_pass3.h

@@ -1,64 +1,20 @@
 #define INIT    
-#define FINI
-#define EMIT(A)     \
-    if(box86_dynarec_dump) {dynarec_log(LOG_NONE, "\t%08x\t%s\n", (A), arm_print(A));} \
-    *(uint32_t*)(dyn->block) = A;       \
-    dyn->block += 4; dyn->arm_size += 4;\
-    dyn->insts[ninst].size2 += 4
+#define FINI        \
+    if(ninst)       \
+        addInst(dyn->instsize, &dyn->insts_size, dyn->insts[ninst].x86.size, dyn->insts[ninst].size/4); \
+    addInst(dyn->instsize, &dyn->insts_size, 0, 0);
+#define EMIT(A)                                         \
+    do{                                                 \
+        if(box86_dynarec_dump) print_opcode(dyn, ninst, (uint32_t)(A)); \
+        if((uintptr_t)dyn->block<(uintptr_t)dyn->next-sizeof(void*))\
+            *(uint32_t*)(dyn->block) = (uint32_t)(A);   \
+        dyn->block += 4; dyn->arm_size += 4;            \
+        dyn->insts[ninst].size2 += 4;                   \
+    }while(0)
 
 #define MESSAGE(A, ...)  if(box86_dynarec_dump) dynarec_log(LOG_NONE, __VA_ARGS__)
 #define NEW_INST        \
-    if(ninst && isInstClean(dyn, ninst)) {                      \
-        dyn->sons_x86[dyn->sons_size] = (uintptr_t)ip;          \
-        dyn->sons_arm[dyn->sons_size] = dyn->block;             \
-        MESSAGE(LOG_DUMP, "----> potential Son here %p/%p\n", (void*)ip, dyn->block);  \
-        ++dyn->sons_size;                                       \
-    }
-
+    if(ninst)                                                   \
+        addInst(dyn->instsize, &dyn->insts_size, dyn->insts[ninst-1].x86.size, dyn->insts[ninst-1].size/4);
 #define INST_EPILOG     
-#define INST_NAME(name) \
-    if(box86_dynarec_dump) {\
-        printf_x86_instruction(my_context->dec, &dyn->insts[ninst].x86, name); \
-        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X", \
-            (box86_dynarec_dump>1)?"\e[32m":"", \
-            (void*)(dyn->arm_start+dyn->insts[ninst].address),  \
-            dyn->insts[ninst].size/4,           \
-            ninst,                              \
-            dyn->insts[ninst].x86.barrier,      \
-            dyn->insts[ninst].x86.state_flags,  \
-            dyn->f.pending,                     \
-            dyn->f.dfnone,                      \
-            dyn->insts[ninst].x86.may_set?"may":"set",          \
-            dyn->insts[ninst].x86.set_flags,    \
-            dyn->insts[ninst].x86.gen_flags,    \
-            dyn->insts[ninst].x86.use_flags,    \
-            dyn->insts[ninst].x86.need_before,  \
-            dyn->insts[ninst].x86.need_after);  \
-        if(dyn->insts[ninst].pred_sz) {         \
-            dynarec_log(LOG_NONE, ", pred=");     \
-            for(int ii=0; ii<dyn->insts[ninst].pred_sz; ++ii)\
-                dynarec_log(LOG_NONE, "%s%d", ii?"/":"", dyn->insts[ninst].pred[ii]);\
-        }                                       \
-        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts>=0)\
-            dynarec_log(LOG_NONE, ", jmp=%d", dyn->insts[ninst].x86.jmp_insts);\
-        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts==-1)\
-            dynarec_log(LOG_NONE, ", jmp=out");\
-        for(int ii=0; ii<24; ++ii) {            \
-            switch(dyn->insts[ninst].n.neoncache[ii].t) {    \
-                case NEON_CACHE_ST_D: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;            \
-                case NEON_CACHE_ST_F: dynarec_log(LOG_NONE, " S%d:%s", (ii+8)*2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;        \
-                case NEON_CACHE_MM: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;              \
-                case NEON_CACHE_XMMW: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;  \
-                case NEON_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;  \
-                case NEON_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;             \
-                case NEON_CACHE_NONE:           \
-                default:    break;              \
-            }                                   \
-        }                                       \
-        if(dyn->n.stack || dyn->insts[ninst].n.stack_next || dyn->insts[ninst].n.x87stack)     \
-            dynarec_log(LOG_NONE, " X87:%d/%d(+%d/-%d)%d", dyn->n.stack, dyn->insts[ninst].n.stack_next, dyn->insts[ninst].n.stack_push, dyn->insts[ninst].n.stack_pop, dyn->insts[ninst].n.x87stack); \
-        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)                     \
-            dynarec_log(LOG_NONE, " %s:%d/%d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);   \
-        dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");                       \
-    }
-
+#define INST_NAME(name) inst_name_pass3(dyn, ninst, name)

src/dynarec/dynarec_arm_private.h

@@ -5,6 +5,7 @@
 
 typedef struct x86emu_s x86emu_t;
 typedef struct dynablock_s dynablock_t;
+typedef struct instsize_s instsize_t;
 
 #define BARRIER_MAYBE   8
 
@@ -68,7 +69,7 @@ typedef struct instruction_arm_s {
     int                 pred_sz;    // size of predecessor list
     int                 *pred;      // predecessor array
     uintptr_t           mark, mark2, mark3;
-    uintptr_t           markf;
+    uintptr_t           markf, markf2;
     uintptr_t           markseg;
     uintptr_t           marklock;
     int                 pass2choice;// value for choices that are fixed on pass2 for pass3
@@ -90,16 +91,23 @@ typedef struct dynarec_arm_s {
     void*               block;      // memory pointer where next instruction is emited
     uintptr_t           arm_start;  // start of the arm code
     int                 arm_size;   // size of emitted arm code
+    uintptr_t           jmp_next;   // address of the jump_next address
     flagcache_t         f;
     neoncache_t         n;          // cache for the 8..31 double reg from fpu, plus x87 stack delta
     uintptr_t*          next;       // variable array of "next" jump address
     int                 next_sz;
     int                 next_cap;
-    uintptr_t*          sons_x86;   // the x86 address of potential dynablock sons
-    void**              sons_arm;   // the arm address of potential dynablock sons
-    int                 sons_size;  // number of potential dynablock sons
     int*                predecessor;// single array of all predecessor
     dynablock_t*        dynablock;
+    instsize_t*         instsize;
+    size_t              insts_size; // size of the instruction size array (calculated)
+    uint8_t             smread;    // for strongmem model emulation
+    uint8_t             smwrite;    // for strongmem model emulation
+    uintptr_t           forward;    // address of the last end of code while testing forward
+    uintptr_t           forward_to; // address of the next jump to (to check if everything is ok)
+    int32_t             forward_size;   // size at the forward point
+    int                 forward_ninst;  // ninst at the forward point
+    uint8_t             always_test;
 } dynarec_arm_t;
 
 void add_next(dynarec_arm_t *dyn, uintptr_t addr);
@@ -107,4 +115,6 @@ uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr);
 int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n);
 int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n);
 
+void CreateJmpNext(void* addr, void* next);
+
 #endif //__DYNAREC_ARM_PRIVATE_H_

src/elfs/elfload_dump.c

@@ -235,12 +235,12 @@ void DumpMainHeader(Elf32_Ehdr *header, elfheader_t *h)
         printf_dump(LOG_NEVER, "ELF Dump ==========\n");
 
         printf_dump(LOG_NEVER, "ELF Dump PEntries (%d)\n", h->numSHEntries);
-        for (int i=0; i<h->numPHEntries; ++i)
+        for (uint16_t i=0; i<h->numPHEntries; ++i)
             printf_dump(LOG_NEVER, "  PHEntry %04d : %s\n", i, DumpPHEntry(h->PHEntries+i));
         printf_dump(LOG_NEVER, "ELF Dump PEntries ====\n");
 
         printf_dump(LOG_NEVER, "ELF Dump Sections (%d)\n", h->numSHEntries);
-        for (int i=0; i<h->numSHEntries; ++i)
+        for (uint16_t i=0; i<h->numSHEntries; ++i)
             printf_dump(LOG_NEVER, "  Section %04d : %s\n", i, DumpSection(h->SHEntries+i, h->SHStrTab));
         printf_dump(LOG_NEVER, "ELF Dump Sections ====\n");
     }
@@ -251,7 +251,7 @@ void DumpSymTab(elfheader_t *h)
     if(box86_dump && h->SymTab) {
         const char* name = ElfName(h);
         printf_dump(LOG_NEVER, "ELF Dump SymTab(%d)=\n", h->numSymTab);
-        for (int i=0; i<h->numSymTab; ++i)
+        for (uint32_t i=0; i<h->numSymTab; ++i)
             printf_dump(LOG_NEVER, "  %s:SymTab[%d] = \"%s\", value=%p, size=%d, info/other=%d/%d index=%d\n", name, 
                 i, h->StrTab+h->SymTab[i].st_name, (void*)h->SymTab[i].st_value, h->SymTab[i].st_size,
                 h->SymTab[i].st_info, h->SymTab[i].st_other, h->SymTab[i].st_shndx);
@@ -263,7 +263,7 @@ void DumpDynamicSections(elfheader_t *h)
 {
     if(box86_dump && h->Dynamic) {
         printf_dump(LOG_NEVER, "ELF Dump Dynamic(%d)=\n", h->numDynamic);
-        for (int i=0; i<h->numDynamic; ++i)
+        for (uint32_t i=0; i<h->numDynamic; ++i)
             printf_dump(LOG_NEVER, "  Dynamic %04d : %s\n", i, DumpDynamic(h->Dynamic+i));
         printf_dump(LOG_NEVER, "ELF Dump Dynamic=====\n");
     }
@@ -274,7 +274,7 @@ void DumpDynSym(elfheader_t *h)
     if(box86_dump && h->DynSym) {
         const char* name = ElfName(h);
         printf_dump(LOG_NEVER, "ELF Dump DynSym(%d)=\n", h->numDynSym);
-        for (int i=0; i<h->numDynSym; ++i) {
+        for (uint32_t i=0; i<h->numDynSym; ++i) {
             int version = h->VerSym?((Elf32_Half*)((uintptr_t)h->VerSym+h->delta))[i]:-1;
             printf_dump(LOG_NEVER, "  %s:DynSym[%d] = %s\n", name, i, DumpSym(h, h->DynSym+i, version));
         }
@@ -286,7 +286,7 @@ void DumpDynamicNeeded(elfheader_t *h)
 {
     if(box86_dump && h->DynStrTab) {
         printf_dump(LOG_NEVER, "ELF Dump DT_NEEDED=====\n");
-        for (int i=0; i<h->numDynamic; ++i)
+        for (uint32_t i=0; i<h->numDynamic; ++i)
             if(h->Dynamic[i].d_tag==DT_NEEDED) {
                 printf_dump(LOG_NEVER, "  Needed : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
             }
@@ -298,7 +298,7 @@ void DumpDynamicRPath(elfheader_t *h)
 {
     if(box86_dump && h->DynStrTab) {
         printf_dump(LOG_NEVER, "ELF Dump DT_RPATH/DT_RUNPATH=====\n");
-        for (int i=0; i<h->numDynamic; ++i) {
+        for (uint32_t i=0; i<h->numDynamic; ++i) {
             if(h->Dynamic[i].d_tag==DT_RPATH) {
                 printf_dump(LOG_NEVER, "   RPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
             }

src/elfs/elfloader.c

@@ -1,4 +1,4 @@
-#define _GNU_SOURCE
+    #define _GNU_SOURCE
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -355,6 +355,7 @@ int LoadElfMemory(FILE* f, box86context_t* context, elfheader_t* head)
 
 int ReloadElfMemory(FILE* f, box86context_t* context, elfheader_t* head)
 {
+    (void)context;
     for (int i=0; i<head->numPHEntries; ++i) {
         if(head->PHEntries[i].p_type == PT_LOAD) {
             Elf32_Phdr * e = &head->PHEntries[i];
@@ -451,8 +452,15 @@ int RelocateElfREL(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
         const char* vername = GetSymbolVersion(head, version);
         const char* defver = GetDefaultVersion((bind==STB_WEAK)?my_context->weakdefver:my_context->globaldefver, symname);
         if(bind==STB_LOCAL) {
-            offs = sym->st_value + head->delta;
-            end = offs + sym->st_size;
+            if(!symname || !symname[0]) {
+                offs = sym->st_value + head->delta;
+                end = offs + sym->st_size;
+            } else {
+                if(!offs && !end && local_maplib)
+                    GetLocalSymbolStartEnd(local_maplib, symname, &offs, &end, head, version, vername);
+                if(!offs && !end)
+                    GetLocalSymbolStartEnd(maplib, symname, &offs, &end, head, version, vername);
+            }
         } else {
             // this is probably very very wrong. A proprer way to get reloc need to be writen, but this hack seems ok for now
             // at least it work for half-life, unreal, ut99, zsnes, Undertale, ColinMcRae Remake, FTL, ShovelKnight...
@@ -576,8 +584,10 @@ int RelocateElfREL(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
                 break;
             case R_386_32:
                 if (!offs) {
-                    printf_log(LOG_NONE, "Error: Symbol sym=%s(%s%s%s/version %d) not found, cannot apply R_386_32 %p (%p) in %s\n", symname, symname, vername?"@":"", vername?vername:"", version, p, *(void**)p, head->name);
-                    ret_ok = 1;
+                        if(strcmp(symname, "__gmon_start__") && strcmp(symname, "data_start") && strcmp(symname, "__data_start")) {
+                            printf_log(LOG_NONE, "Error: Symbol sym=%s(%s%s%s/version %d) not found, cannot apply R_386_32 %p (%p) in %s\n", symname, symname, vername?"@":"", vername?vername:"", version, p, *(void**)p, head->name);
+                            ret_ok = 1;
+                        }
                 } else {
                     printf_dump(LOG_NEVER, "Apply %s R_386_32 %p with sym=%s (ver=%d/%s) (%p -> %p)\n", (bind==STB_LOCAL)?"Local":((bind==STB_WEAK)?"Weak":"Global"), p, symname, version, vername?vername:"(none)", *(void**)p, (void*)(offs+*(uint32_t*)p));
                     *p += offs;
@@ -586,7 +596,7 @@ int RelocateElfREL(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
             case R_386_TLS_TPOFF:
                 // Negated offset in static TLS block
                 {
-                    if(!symname || !symname[0]) {
+                    if(!symname || !symname[0] || bind==STB_LOCAL) {
                         h_tls = head;
                         offs = sym->st_value;
                     } else {
@@ -656,12 +666,12 @@ int RelocateElfREL(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
                         printf_log(LOG_NONE, "Error: Symbol %s not found, cannot apply R_386_TLS_DTPOFF32 %p (%p) in %s\n", symname, p, *(void**)p, head->name);
                     }
                 } else {
-                    if(h_tls)
-                        offs = sym->st_value;
+                    if(!symname || !symname[0]) {
+                        offs = (uintptr_t)((intptr_t)(head->tlsaddr + head->delta) - (intptr_t)offs);    // negative offset
+                    }
                     if(p) {
-                        int tlsoffset = offs;    // it's not an offset in elf memory
-                        printf_dump(LOG_NEVER, "Apply %s R_386_TLS_DTPOFF32 %p with sym=%s (%p -> %p)\n", (bind==STB_LOCAL)?"Local":((bind==STB_WEAK)?"Weak":"Global"), p, symname, (void*)tlsoffset, (void*)offs);
-                        *p = tlsoffset;
+                        printf_dump(LOG_NEVER, "Apply %s R_386_TLS_DTPOFF32 %p with sym=%s (ver=%d/%s) (%zd -> %zd)\n", (bind==STB_LOCAL)?"Local":((bind==STB_WEAK)?"Weak":"Global"), p, symname, version, vername?vername:"(none)", (intptr_t)*p, (intptr_t)offs);
+                        *p = offs;
                     } else {
                         printf_log(LOG_NONE, "Warning, Symbol %s found, but R_386_TLS_DTPOFF32 Slot Offset is NULL \n", symname);
                     }
@@ -693,8 +703,15 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
         const char* vername = GetSymbolVersion(head, version);
         const char* defver = GetDefaultVersion((bind==STB_WEAK)?my_context->weakdefver:my_context->globaldefver, symname);
         if(bind==STB_LOCAL) {
-            offs = sym->st_value + head->delta;
-            end = offs + sym->st_size;
+            if(!symname || !symname[0]) {
+                offs = sym->st_value + head->delta;
+                end = offs + sym->st_size;
+            } else {
+                if(!offs && !end && local_maplib)
+                    GetLocalSymbolStartEnd(local_maplib, symname, &offs, &end, head, version, vername);
+                if(!offs && !end)
+                    GetLocalSymbolStartEnd(maplib, symname, &offs, &end, head, version, vername);
+            }
         } else {
             // this is probably very very wrong. A proprer way to get reloc need to be writen, but this hack seems ok for now
             // at least it work for half-life, unreal, ut99, zsnes, Undertale, ColinMcRae Remake, FTL, ShovelKnight...
@@ -827,7 +844,7 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
             case R_386_TLS_TPOFF:
                 // Negated offset in static TLS block
                 {
-                    if(!symname || !symname[0]) {
+                    if(!symname || !symname[0] || bind==STB_LOCAL) {
                         h_tls = head;
                         offs = sym->st_value;
                     } else {
@@ -897,8 +914,9 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
                         printf_log(LOG_NONE, "Error: Symbol %s not found, cannot apply R_386_TLS_DTPOFF32 %p (%p) in %s\n", symname, p, *(void**)p, head->name);
                     }
                 } else {
-                    if(h_tls)
-                        offs = sym->st_value;
+                   if(!symname || !symname[0]) {
+                        offs = (uintptr_t)((intptr_t)(head->tlsaddr + head->delta) - (intptr_t)offs);    // negative offset
+                    }
                     if(p) {
                         int tlsoffset = offs;    // it's not an offset in elf memory
                         printf_dump(LOG_NEVER, "Apply %s R_386_TLS_DTPOFF32 %p with sym=%s (%p -> %p)\n", (bind==STB_LOCAL)?"Local":((bind==STB_WEAK)?"Weak":"Global"), p, symname, (void*)tlsoffset, (void*)offs);
@@ -952,7 +970,7 @@ int RelocateElfPlt(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
                 return -1;
         }
         if(need_resolver) {
-            if(pltResolver==~0) {
+            if(pltResolver==~(uintptr_t)0) {
                 pltResolver = AddBridge(my_context->system, vFEv, PltResolver, 0, "(PltResolver)");
             }
             if(head->pltgot) {
@@ -970,7 +988,7 @@ int RelocateElfPlt(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
     return 0;
 }
 
-void CalcStack(elfheader_t* elf, uint32_t* stacksz, int* stackalign)
+void CalcStack(elfheader_t* elf, uint32_t* stacksz, uint32_t* stackalign)
 {
     if(*stacksz < elf->stacksz)
         *stacksz = elf->stacksz;
@@ -981,7 +999,7 @@ void CalcStack(elfheader_t* elf, uint32_t* stacksz, int* stackalign)
 Elf32_Sym* GetFunction(elfheader_t* h, const char* name)
 {
     // TODO: create a hash on named to avoid this loop
-    for (int i=0; i<h->numSymTab; ++i) {
+    for (uint32_t i=0; i<h->numSymTab; ++i) {
         int type = ELF32_ST_TYPE(h->SymTab[i].st_info);
         if(type==STT_FUNC) {
             const char * symname = h->StrTab+h->SymTab[i].st_name;
@@ -995,7 +1013,7 @@ Elf32_Sym* GetFunction(elfheader_t* h, const char* name)
 
 Elf32_Sym* GetElfObject(elfheader_t* h, const char* name)
 {
-    for (int i=0; i<h->numSymTab; ++i) {
+    for (uint32_t i=0; i<h->numSymTab; ++i) {
         int type = ELF32_ST_TYPE(h->SymTab[i].st_info);
         if(type==STT_OBJECT) {
             const char * symname = h->StrTab+h->SymTab[i].st_name;
@@ -1017,6 +1035,7 @@ uintptr_t GetFunctionAddress(elfheader_t* h, const char* name)
 
 uintptr_t GetEntryPoint(lib_t* maplib, elfheader_t* h)
 {
+    (void)maplib;
     uintptr_t ep = h->entrypoint + h->delta;
     printf_log(LOG_DEBUG, "Entry Point is %p\n", (void*)ep);
     if(box86_dump) {
@@ -1063,7 +1082,7 @@ void AddSymbols(lib_t *maplib, kh_mapsymbols_t* mapsymbols, kh_mapsymbols_t* wea
             printf_dump(LOG_NEVER, "Adding %s Default Version \"%s\" for Symbol\"%s\"\n", (bind==STB_WEAK)?"Weak":"Global", vername, symname);
         }
         if((type==STT_OBJECT || type==STT_FUNC || type==STT_COMMON || type==STT_TLS  || type==STT_NOTYPE) 
-        && (vis==STV_DEFAULT || vis==STV_PROTECTED) && (h->SymTab[i].st_shndx!=0)) {
+        && (vis==STV_DEFAULT || vis==STV_PROTECTED || (vis==STV_HIDDEN && bind==STB_LOCAL)) && (h->SymTab[i].st_shndx!=0)) {
             if(strstr(symname, "@@")) {
                 char symnameversionned[strlen(symname)+1];
                 strcpy(symnameversionned, symname);
@@ -1110,7 +1129,7 @@ void AddSymbols(lib_t *maplib, kh_mapsymbols_t* mapsymbols, kh_mapsymbols_t* wea
         int type = ELF32_ST_TYPE(h->DynSym[i].st_info);
         int vis = h->DynSym[i].st_other&0x3;
         if((type==STT_OBJECT || type==STT_FUNC || type==STT_COMMON || type==STT_TLS  || type==STT_NOTYPE) 
-        && (vis==STV_DEFAULT || vis==STV_PROTECTED) && (h->DynSym[i].st_shndx!=0 && h->DynSym[i].st_shndx<=65521)) {
+        && (vis==STV_DEFAULT || vis==STV_PROTECTED || (vis==STV_HIDDEN && bind==STB_LOCAL)) && (h->DynSym[i].st_shndx!=0 && h->DynSym[i].st_shndx<=65521)) {
             uintptr_t offs = (type==STT_TLS)?h->DynSym[i].st_value:(h->DynSym[i].st_value + h->delta);
             size_t sz = h->DynSym[i].st_size;
             int version = h->VerSym?((Elf32_Half*)((uintptr_t)h->VerSym+h->delta))[i]:-1;
@@ -1156,7 +1175,7 @@ int LoadNeededLibs(elfheader_t* h, lib_t* maplib, needed_libs_t* neededlibs, lib
 {
     DumpDynamicRPath(h);
     // update RPATH first
-    for (int i=0; i<h->numDynamic; ++i)
+    for (uint32_t i=0; i<h->numDynamic; ++i)
         if(h->Dynamic[i].d_tag==DT_RPATH || h->Dynamic[i].d_tag==DT_RUNPATH) {
             char *rpathref = h->DynStrTab+h->delta+h->Dynamic[i].d_un.d_val;
             char* rpath = rpathref;
@@ -1213,12 +1232,12 @@ int LoadNeededLibs(elfheader_t* h, lib_t* maplib, needed_libs_t* neededlibs, lib
 
     DumpDynamicNeeded(h);
     int cnt = 0;
-    for (int i=0; i<h->numDynamic; ++i)
+    for (uint32_t i=0; i<h->numDynamic; ++i)
         if(h->Dynamic[i].d_tag==DT_NEEDED)
             ++cnt;
     const char* nlibs[cnt];
     int j=0;
-    for (int i=0; i<h->numDynamic; ++i)
+    for (uint32_t i=0; i<h->numDynamic; ++i)
         if(h->Dynamic[i].d_tag==DT_NEEDED)
             nlibs[j++] = h->DynStrTab+h->delta+h->Dynamic[i].d_un.d_val;
 
@@ -1235,7 +1254,7 @@ int ElfCheckIfUseTCMallocMinimal(elfheader_t* h)
 {
     if(!h)
         return 0;
-    for (int i=0; i<h->numDynamic; ++i)
+    for (uint32_t i=0; i<h->numDynamic; ++i)
         if(h->Dynamic[i].d_tag==DT_NEEDED) {
             char *needed = h->DynStrTab+h->delta+h->Dynamic[i].d_un.d_val;
             if(!strcmp(needed, "libtcmalloc_minimal.so.4")) // tcmalloc needs to be the 1st lib loaded
@@ -1270,6 +1289,34 @@ void MarkElfInitDone(elfheader_t* h)
     if(h)
         h->init_done = 1;
 }
+void RunElfInitPltResolver(elfheader_t* h, x86emu_t *emu)
+{
+    if(!h || h->init_done)
+        return;
+
+    uintptr_t p = h->initentry + h->delta;
+    h->init_done = 1;
+    for(int i=0; i<h->neededlibs->size; ++i) {
+        library_t *lib = h->neededlibs->libs[i];
+        elfheader_t *lib_elf = GetElf(lib);
+        if(lib_elf)
+            RunElfInitPltResolver(lib_elf, emu);
+    }
+    printf_log(LOG_DEBUG, "Calling Init for %s %p\n", ElfName(h), (void*)p);
+    if(h->initentry)
+        RunSafeFunction(my_context, p, 3, my_context->argc, my_context->argv, my_context->envv);
+    printf_log(LOG_DEBUG, "Done Init for %s\n", ElfName(h));
+    // and check init array now
+    Elf32_Addr *addr = (Elf32_Addr*)(h->initarray + h->delta);
+    for (int i=0; i<h->initarray_sz; ++i) {
+        printf_log(LOG_DEBUG, "Calling Init[%d] for %s %p\n", i, ElfName(h), (void*)addr[i]);
+        RunSafeFunction(my_context, (uintptr_t)addr[i], 3, my_context->argc, my_context->argv, my_context->envv);
+    }
+
+    h->fini_done = 0;   // can be fini'd now (in case it was re-inited)
+    printf_log(LOG_DEBUG, "All Init Done for %s\n", ElfName(h));
+    return;
+}
 void RunElfInit(elfheader_t* h, x86emu_t *emu)
 {
     if(!h || h->init_done)
@@ -1406,7 +1453,7 @@ const char* FindNearestSymbolName(elfheader_t* h, void* p, uintptr_t* start, uin
     if(!h)
         return ret;
 
-    for (int i=0; (i<h->numSymTab) && (distance!=0); ++i) {   
+    for (uint32_t i=0; (i<h->numSymTab) && (distance!=0); ++i) {
         const char * symname = h->StrTab+h->SymTab[i].st_name;
         uintptr_t offs = h->SymTab[i].st_value + h->delta;
 
@@ -1419,7 +1466,7 @@ const char* FindNearestSymbolName(elfheader_t* h, void* p, uintptr_t* start, uin
             }
         }
     }
-    for (int i=0; (i<h->numDynSym) && (distance!=0); ++i) {   
+    for (uint32_t i=0; (i<h->numDynSym) && (distance!=0); ++i) {
         const char * symname = h->DynStr+h->DynSym[i].st_name;
         uintptr_t offs = h->DynSym[i].st_value + h->delta;
 
@@ -1540,7 +1587,7 @@ void* ElfSetBrk(void* newbrk)
         printf_log(LOG_NONE, "Error, using brk on multi-part memory elf is not allowed for now");
     } else {
         printf_log(LOG_DEBUG, "brk change, start=%p with reserve=0x%x, was 0x%x long, will be 0x%x now => %p\n", my_context->brk, h->reserve, my_context->brksz, added, my_context->brk+added);
-        if(added>0 && added<h->reserve) {
+        if(added>0 && (uint32_t)added<h->reserve) {
             my_context->brksz = added;
             return newbrk;
         }
@@ -1560,16 +1607,12 @@ void* ElfSetBrk(void* newbrk)
 }
 
 #ifdef DYNAREC
-dynablocklist_t* GetDynablocksFromAddress(box86context_t *context, uintptr_t addr)
+dynablock_t* GetDynablocksFromAddress(box86context_t *context, uintptr_t addr)
 {
-    // if we are here, the there is not block in standard "space"
-    /*dynablocklist_t* ret = getDBFromAddress(addr);
-    if(ret) {
-        return ret;
-    }*/
+    (void)context;
     if(box86_dynarec_forced) {
         addDBFromAddressRange(addr, 1);
-        return getDB(addr>>DYNAMAP_SHIFT);
+        return getDB(addr);
     }
     //check if address is in an elf... if yes, grant a block (should I warn)
     Dl_info info;
@@ -1673,7 +1716,7 @@ EXPORT int my_dl_iterate_phdr(x86emu_t *emu, void* F, void *data) {
 void ResetSpecialCaseMainElf(elfheader_t* h)
 {
     Elf32_Sym *sym = NULL;
-     for (int i=0; i<h->numDynSym; ++i) {
+     for (uint32_t i=0; i<h->numDynSym; ++i) {
         if(h->DynSym[i].st_info == 17) {
             sym = h->DynSym+i;
             const char * symname = h->DynStr+sym->st_name;
@@ -1761,6 +1804,7 @@ typedef struct search_symbol_s{
 } search_symbol_t;
 int dl_iterate_phdr_findsymbol(struct dl_phdr_info* info, size_t size, void* data)
 {
+    (void)size;
     search_symbol_t* s = (search_symbol_t*)data;
 
     for(int j = 0; j<info->dlpi_phnum; ++j) {
@@ -1825,7 +1869,7 @@ void* GetNativeSymbolUnversionned(void* lib, const char* name)
     return s.addr;
 }
 
-uintptr_t pltResolver = ~0;
+uintptr_t pltResolver = ~(uintptr_t)0;
 EXPORT void PltResolver(x86emu_t* emu)
 {
     uintptr_t addr = Pop32(emu);
@@ -1857,6 +1901,12 @@ EXPORT void PltResolver(x86emu_t* emu)
         emu->quit = 1;
         return;
     } else {
+        elfheader_t* sym_elf = FindElfAddress(my_context, offs);
+        if(sym_elf && sym_elf!=my_context->elfs[0] && !sym_elf->init_done) {
+            printf_dump(LOG_DEBUG, "symbol %s from %s but elf not initialized yet, run Init now (from %s)\n", symname, ElfName(sym_elf), ElfName(h));
+            RunElfInitPltResolver(sym_elf, emu);
+        }
+
         if(p) {
             printf_dump(LOG_DEBUG, "            Apply %s R_386_JMP_SLOT %p with sym=%s(ver %d: %s%s%s) (%p -> %p / %s)\n", (bind==STB_LOCAL)?"Local":((bind==STB_WEAK)?"Weak":"Global"), p, symname, version, symname, vername?"@":"", vername?vername:"",*(void**)p, (void*)offs, ElfName(FindElfAddress(my_context, offs)));
             *p = offs;

src/elfs/elfloader_private.h

@@ -1,10 +1,6 @@
 #ifndef __ELFLOADER_PRIVATE_H_
 #define __ELFLOADER_PRIVATE_H_
 
-#ifdef DYNAREC
-typedef struct dynablocklist_s dynablocklist_t;
-#endif
-
 typedef struct library_s library_t;
 typedef struct needed_libs_s needed_libs_t;
 typedef struct kh_mapsymbols_s kh_mapsymbols_t;
@@ -15,21 +11,21 @@ typedef struct kh_mapsymbols_s kh_mapsymbols_t;
 struct elfheader_s {
     char*       name;
     char*       path;   // Resolved path to file
-    int         numPHEntries;
+    uint16_t    numPHEntries;
     Elf32_Phdr  *PHEntries;
-    int         numSHEntries;
+    uint16_t    numSHEntries;
     Elf32_Shdr  *SHEntries;
     int         SHIdx;
     int         numSST;
     char*       SHStrTab;
     char*       StrTab;
     Elf32_Sym*  SymTab;
-    int         numSymTab;
+    uint32_t    numSymTab;
     char*       DynStr;
     Elf32_Sym*  DynSym;
-    int         numDynSym;
+    uint32_t    numDynSym;
     Elf32_Dyn*  Dynamic;
-    int         numDynamic;
+    uint32_t    numDynamic;
     char*       DynStrTab;
     int         szDynStrTab;
     Elf32_Half* VerSym;
@@ -73,15 +69,15 @@ struct elfheader_s {
 
     uintptr_t   paddr;
     uintptr_t   vaddr;
-    int         align;
+    uint32_t    align;
     uint32_t    memsz;
     uint32_t    reserve;
     uint32_t    stacksz;
-    int         stackalign;
+    uint32_t    stackalign;
     uintptr_t   tlsaddr;
     uint32_t    tlssize;
     uint32_t    tlsfilesize;
-    int         tlsalign;
+    uint32_t    tlsalign;
 
     int32_t     tlsbase;    // the base of the tlsdata in the global tlsdata (always negative)
 
@@ -116,4 +112,4 @@ struct elfheader_s {
 
 elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec);
 
-#endif //__ELFLOADER_PRIVATE_H_
+#endif //__ELFLOADER_PRIVATE_H_

src/elfs/elfparser.c

@@ -43,7 +43,7 @@ int FindSection(Elf32_Shdr *s, int n, char* SHStrTab, const char* name)
     return 0;
 }
 
-void LoadNamedSection(FILE *f, Elf32_Shdr *s, int size, char* SHStrTab, const char* name, const char* clearname, uint32_t type, void** what, int* num)
+void LoadNamedSection(FILE *f, Elf32_Shdr *s, int size, char* SHStrTab, const char* name, const char* clearname, uint32_t type, void** what, uint32_t* num)
 {
     int n = FindSection(s, size, SHStrTab, name);
     printf_dump(LOG_DEBUG, "Loading %s (idx = %d)\n", clearname, n);
@@ -136,7 +136,7 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
             h->numSHEntries = section.sh_size;
         }
         // now read all section headers
-        printf_dump(LOG_DEBUG, "Read %d Section header\n", h->numSHEntries);
+        printf_dump(LOG_DEBUG, "Read %u Section header\n", h->numSHEntries);
         h->SHEntries = (Elf32_Shdr*)box_calloc(h->numSHEntries, sizeof(Elf32_Shdr));
         fseeko64(f, header.e_shoff ,SEEK_SET);
         if(fread(h->SHEntries, sizeof(Elf32_Shdr), h->numSHEntries, f)!=h->numSHEntries) {
@@ -188,7 +188,7 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
         // grab DT_REL & DT_RELA stuffs
         // also grab the DT_STRTAB string table
         {
-            for (int i=0; i<h->numDynamic; ++i) {
+            for (uint32_t i=0; i<h->numDynamic; ++i) {
                 Elf32_Dyn d = h->Dynamic[i];
                 Elf32_Word val = d.d_un.d_val;
                 Elf32_Addr ptr = d.d_un.d_ptr;

src/emu/modrm.h

@@ -1,3 +1,25 @@
+#include <string.h>
+
+#define F8      *(uint8_t*)(addr++)
+#define F8S     *(int8_t*)(addr++)
+#define F16     *(uint16_t*)(addr+=2, addr-2)
+#define F16S    *(int16_t*)(addr+=2, addr-2)
+#define F32     *(uint32_t*)(addr+=4, addr-4)
+#define F32S    *(int32_t*)(addr+=4, addr-4)
+#define F32S64  (uint64_t)(int64_t)F32S
+#define F64     *(uint64_t*)(addr+=8, addr-8)
+#define F64S    *(int64_t*)(addr+=8, addr-8)
+#define PK(a)   *(uint8_t*)(addr+a)
+#ifdef DYNAREC
+#define STEP if(step) return 0;
+#define STEP2 if(step) {R_EIP = addr; return 0;}
+#define STEP3 if(*step) (*step)++;
+#else
+#define STEP
+#define STEP2
+#define STEP3
+#endif
+
 // ModRM utilities macros
 #define getecommon(A, T) \
     if(!(nextop&0xC0)) { \
@@ -97,6 +119,17 @@
     if((nextop&0xC0)==0xC0) { \
         A = (reg32_t*)&emu->regs[(nextop&3)].byte[((nextop&0x4)>>2)]; \
     } else getecommon(A, reg32_t)
+#define testeb(A, O) \
+    if((nextop&0xC0)==0xC0) { \
+        A = (reg32_t*)&emu->regs[(nextop&3)].byte[((nextop&0x4)>>2)]; \
+    } else { \
+        reg32_t* ret; \
+        getecommono(ret, reg32_t, O);\
+        test->memsize = 1; \
+        test->memaddr = (uintptr_t)ret;\
+        test->mem[0] = ret->byte[0];\
+        A = (reg32_t*)test->mem;\
+    }
 #define getebo(A, O)          \
     if((nextop&0xC0)==0xC0) { \
         A = (reg32_t*)&emu->regs[(nextop&3)].byte[((nextop&0x4)>>2)]; \
@@ -109,14 +142,47 @@
     if((nextop&0xC0)==0xC0) { \
         A = &emu->regs[(nextop&7)]; \
     } else getecommono(A, reg32_t, O)
+#define tested(A, SZ, O) \
+    if((nextop&0xC0)==0xC0) { \
+        A = &emu->regs[(nextop&7)]; \
+    } else {\
+        reg32_t* ret; \
+        getecommono(ret, reg32_t, O);\
+        test->memsize = SZ; \
+        test->memaddr = (uintptr_t)ret;\
+        memcpy(test->mem, ret->dword, SZ);\
+        A = (reg32_t*)test->mem;\
+    }
 #define getem(A) \
     if((nextop&0xC0)==0xC0) { \
         A = &emu->mmx[(nextop&7)]; \
     } else getecommon(A, mmx87_regs_t)
+#define testem(A) \
+    if((nextop&0xC0)==0xC0) { \
+        A = &emu->mmx[(nextop&7)]; \
+    } else {\
+        mmx87_regs_t* ret; \
+        getecommon(ret, mmx87_regs_t) \
+        test->memsize = 8; \
+        test->memaddr = (uintptr_t)ret;\
+        memcpy(test->mem, ret->ub, 8);\
+        A = (mmx87_regs_t*)test->mem;\
+    }
 #define getex(A) \
     if((nextop&0xC0)==0xC0) { \
         A = &emu->xmm[(nextop&7)]; \
     } else getecommon(A, sse_regs_t)
+#define testex(A) \
+    if((nextop&0xC0)==0xC0) { \
+        A = &emu->xmm[(nextop&7)]; \
+    } else {\
+        sse_regs_t* ret; \
+        getecommon(ret, sse_regs_t);\
+        test->memsize = 16; \
+        test->memaddr = (uintptr_t)ret;\
+        memcpy(test->mem, ret->ub, 16);\
+        A = (sse_regs_t*)test->mem;\
+    }
 #define getew16(A)  \
     if((nextop&0xC0)==0xC0) { \
         A = &emu->regs[(nextop&7)]; \
@@ -125,16 +191,44 @@
     if((nextop&0xC0)==0xC0) {       \
         A = &emu->regs[(nextop&7)]; \
     } else getecommon16o(A, reg32_t, O)
+#define testtew16(A, O)                 \
+    if((nextop&0xC0)==0xC0) {           \
+        A = &emu->regs[(nextop&7)];     \
+    } else {                            \
+        reg32_t* ret;                   \
+        getecommon16o(A, reg32_t, O);   \
+        test->memsize = 2;              \
+        test->memaddr = (uintptr_t)ret; \
+        *(uint16_t*)test->mem = ret->word[0];\
+        A = (reg32_t*)test->mem;        \
+    }
 
 // Macros for ModR/M gets
+#ifdef TEST_INTERPRETER
+#define GET_EB      testeb(oped, 0)
+#define GET_ED      tested(oped, 4, 0)
+#define GET_ED8     tested(oped, 8, 0)
+#define GET_EDT     tested(oped, 10, 0)
+#define GET_ED_OFFS(o) tested(oped, 4, o)
+#define GET_EB_OFFS(o) testeb(oped, o)
+#define GET_EM      testem(opem)
+#define GET_EX      testex(opex)
+#define GET_EW16    testew16(oped, 0)
+#define GET_EW16_OFFS(o)    testtew16(oped, o)
+#else
 #define GET_EB      geteb(oped)
 #define GET_ED      geted(oped)
+#define GET_ED8     geted(oped)
+#define GET_EDT     geted(oped)
 #define GET_ED_OFFS(o) getedo(oped, o)
 #define GET_EB_OFFS(o) getebo(oped, o)
 #define GET_EM      getem(opem)
 #define GET_EX      getex(opex)
 #define GET_EW16    getew16(oped)
 #define GET_EW16_OFFS(o)    getew16o(oped, o)
+#endif
+#define GET_ED_     geted(oped)
+#define GET_EW16_   getew16(oped)
 #define EB          oped
 #define ED          oped
 #define EM          opem

src/emu/run66.h

@@ -1,555 +0,0 @@
-    opcode = F8;
-    goto *opcodes66[opcode];
-
-    #define GO(B, OP)                       \
-    _66_##B##_0: \
-        nextop = F8;               \
-        GET_EB;             \
-        EB->byte[0] = OP##8(emu, EB->byte[0], GB);  \
-        NEXT;                              \
-    _66_##B##_1:                               \
-        nextop = F8;                \
-        GET_EW;                   \
-        EW->word[0] = OP##16(emu, EW->word[0], GW.word[0]); \
-        NEXT;                              \
-    _66_##B##_2: \
-        nextop = F8;               \
-        GET_EB;                   \
-        GB = OP##8(emu, GB, EB->byte[0]); \
-        NEXT;                              \
-    _66_##B##_3:                               \
-        nextop = F8;                \
-        GET_EW;                   \
-        GW.word[0] = OP##16(emu, GW.word[0], EW->word[0]); \
-        NEXT;                              \
-    _66_##B##_4: \
-        R_AL = OP##8(emu, R_AL, F8); \
-        NEXT;                              \
-    _66_##B##_5:                               \
-        R_AX = OP##16(emu, R_AX, F16); \
-        NEXT;
-
-    GO(0x00, add)                   /* ADD 0x01 ~> 0x05 */
-    GO(0x08, or)                    /*  OR 0x09 ~> 0x0D */
-    GO(0x10, adc)                   /* ADC 0x11 ~> 0x15 */
-    GO(0x18, sbb)                   /* SBB 0x19 ~> 0x1D */
-    GO(0x20, and)                   /* AND 0x21 ~> 0x25 */
-    GO(0x28, sub)                   /* SUB 0x29 ~> 0x2D */
-    GO(0x30, xor)                   /* XOR 0x31 ~> 0x35 */
-    //GO(0x38, cmp)                   /* CMP 0x39 ~> 0x3D */
-    #undef GO
-    _66_0x06:                      /* PUSH ES */
-        Push16(emu, emu->segs[_ES]);
-        NEXT;
-    _66_0x07:                      /* POP ES */
-        emu->segs[_ES] = Pop16(emu);    // no check, no use....
-        emu->segs_serial[_ES] = 0;
-        NEXT;
-
-    _66_0x0F:                      /* 66 0f prefix */
-        emu->old_ip = R_EIP;
-        R_EIP = ip-2;   // don't count 66 0F yet
-        Run660F(emu); // implemented in Run660f.c
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-
-    _66_0x1E:                      /* PUSH DS */
-        Push16(emu, emu->segs[_DS]);
-        NEXT;
-    _66_0x1F:                      /* POP DS */
-        emu->segs[_DS] = Pop16(emu);    // no check, no use....
-        emu->segs_serial[_DS] = 0;
-        NEXT;
-        
-    _66_0x26:                      /* ES: */
-        // ignored
-        opcode = F8;
-        goto *opcodes66[opcode];
-    _66_0x2E:                      /* CS: */
-        // ignored
-        opcode = F8;
-        goto *opcodes66[opcode];
-    _66_0x36:                      /* SS: */
-        // ignored
-        opcode = F8;
-        goto *opcodes66[opcode];
-
-    _66_0x39:
-        nextop = F8;
-        GET_EW;
-        cmp16(emu, EW->word[0], GW.word[0]);
-        NEXT;
-    _66_0x3B:
-        nextop = F8;
-        GET_EW;
-        cmp16(emu, GW.word[0], EW->word[0]);
-        NEXT;
-    _66_0x3D:
-        cmp16(emu, R_AX, F16);
-        NEXT;
-    _66_0x3E:                      /* DS: */
-        // ignored
-        opcode = F8;
-        goto *opcodes66[opcode];
-    
-    _66_0x40:
-    _66_0x41:
-    _66_0x42:
-    _66_0x43:
-    _66_0x44:
-    _66_0x45:
-    _66_0x46:
-    _66_0x47:                              /* INC Reg */
-        tmp8u = opcode&7;
-        emu->regs[tmp8u].word[0] = inc16(emu, emu->regs[tmp8u].word[0]);
-        NEXT;
-    _66_0x48:
-    _66_0x49:
-    _66_0x4A:
-    _66_0x4B:
-    _66_0x4C:
-    _66_0x4D:
-    _66_0x4E:
-    _66_0x4F:                              /* DEC Reg */
-        tmp8u = opcode&7;
-        emu->regs[tmp8u].word[0] = dec16(emu, emu->regs[tmp8u].word[0]);
-        NEXT;
-    _66_0x50:
-    _66_0x51:
-    _66_0x52:
-    _66_0x53:
-    _66_0x54:
-    _66_0x55:
-    _66_0x56:
-    _66_0x57:                              /* PUSH Reg */
-        tmp8u = opcode&7;
-        Push16(emu, emu->regs[tmp8u].word[0]);
-        NEXT;
-    _66_0x58:
-    _66_0x59:
-    _66_0x5A:
-    _66_0x5B:
-    _66_0x5C:                      
-    _66_0x5D:
-    _66_0x5E:
-    _66_0x5F:                              /* POP Reg */
-        tmp8u = opcode&7;
-        emu->regs[tmp8u].word[0] = Pop16(emu);
-        NEXT;
-    _66_0x60:                              /* PUSHA */
-        tmp16u = R_SP;
-        Push16(emu, R_AX);
-        Push16(emu, R_CX);
-        Push16(emu, R_DX);
-        Push16(emu, R_BX);
-        Push16(emu, tmp16u);
-        Push16(emu, R_BP);
-        Push16(emu, R_SI);
-        Push16(emu, R_DI);
-        NEXT;
-    _66_0x61:                              /* POPA */
-        R_DI = Pop16(emu);
-        R_SI = Pop16(emu);
-        R_BP = Pop16(emu);
-        R_ESP+=2;   // POP ESP
-        R_BX = Pop16(emu);
-        R_DX = Pop16(emu);
-        R_CX = Pop16(emu);
-        R_AX = Pop16(emu);
-        NEXT;
-
-    _66_0x64:
-        R_EIP = ip-2;
-        RunFS66(emu, GetFSBaseEmu(emu));
-        ip = R_EIP;
-        STEP;
-        NEXT;
-
-    _66_0x66:
-        goto _0x66; // 0x66 0x66 => can remove one 0x66
-
-    _66_0x68:                       /* PUSH u16 */
-        tmp16u = F16;
-        Push16(emu, tmp16u);
-        NEXT;
-    _66_0x69:                      /* IMUL Gw,Ew,Iw */
-        nextop = F8;
-        GET_EW;
-        tmp16u = F16;
-        GW.word[0] = imul16(emu, EW->word[0], tmp16u);
-        NEXT;
-    _66_0x6A:                      /* PUSH Ib (as signed word) */
-        tmp16s = F8S;
-        Push16(emu, (uint16_t)tmp16s);
-        NEXT;
-    _66_0x6B:                      /* IMUL Gw,Ew,Ib */
-        nextop = F8;
-        GET_EW;
-        tmp16s = F8S;
-        GW.word[0] = imul16(emu, EW->word[0], (uint16_t)tmp16s);
-        NEXT;
-
-    _66_0x81:                              /* GRP3 Ew,Iw */
-    _66_0x83:                              /* GRP3 Ew,Ib */
-        nextop = F8;
-        GET_EW;
-        if(opcode==0x81) 
-            tmp16u = F16;
-        else {
-            tmp16s = F8S;
-            tmp16u = (uint16_t)tmp16s;
-        }
-        switch((nextop>>3)&7) {
-            case 0: EW->word[0] = add16(emu, EW->word[0], tmp16u); break;
-            case 1: EW->word[0] =  or16(emu, EW->word[0], tmp16u); break;
-            case 2: EW->word[0] = adc16(emu, EW->word[0], tmp16u); break;
-            case 3: EW->word[0] = sbb16(emu, EW->word[0], tmp16u); break;
-            case 4: EW->word[0] = and16(emu, EW->word[0], tmp16u); break;
-            case 5: EW->word[0] = sub16(emu, EW->word[0], tmp16u); break;
-            case 6: EW->word[0] = xor16(emu, EW->word[0], tmp16u); break;
-            case 7:               cmp16(emu, EW->word[0], tmp16u); break;
-        }
-        NEXT;
-
-    _66_0x85:                              /* TEST Ew,Gw */
-        nextop = F8;
-        GET_EW;
-        test16(emu, EW->word[0], GW.word[0]);
-        NEXT;
-
-    _66_0x87:                              /* XCHG Ew,Gw */
-        nextop = F8;
-        GET_EW;
-        tmp16u = GW.word[0];
-        GW.word[0] = EW->word[0];
-        EW->word[0] = tmp16u;
-        NEXT;
-
-    _66_0x89:                              /* MOV Ew,Gw */
-        nextop = F8;
-        GET_EW;
-        EW->word[0] = GW.word[0];
-        NEXT;
-
-    _66_0x8B:                              /* MOV Gw,Ew */
-        nextop = F8;
-        GET_EW;
-        GW.word[0] = EW->word[0];
-        NEXT;
-    _66_0x8C:                              /* MOV Ew,Seg */
-        nextop = F8;
-        GET_EW;
-        EW->word[0] = emu->segs[(nextop&0x38)>>3];
-        NEXT;
-    
-    _66_0x8E:                               /* MOV Seg,Ew */
-        nextop = F8;
-        GET_EW;
-        emu->segs[((nextop&0x38)>>3)] = EW->word[0];
-        emu->segs_serial[((nextop&0x38)>>3)] = 0;
-        NEXT;
-    _66_0x8F:                              /* POP Ew */
-        nextop = F8;
-        GET_EW;
-        EW->word[0] = Pop16(emu);
-        NEXT;
-    _66_0x90:                              /* NOP */
-        NEXT;
-
-    _66_0x91:
-    _66_0x92:
-    _66_0x93:
-    _66_0x94:
-    _66_0x95:
-    _66_0x96:
-    _66_0x97:                      /* XCHG reg,EAX */
-        tmp16u = R_AX;
-        R_AX = emu->regs[opcode&7].word[0];
-        emu->regs[opcode&7].word[0] = tmp16u;
-        NEXT;
-
-    _66_0x98:                               /* CBW */
-        emu->regs[_AX].sword[0] = emu->regs[_AX].sbyte[0];
-        NEXT;
-    _66_0x99:                              /* CWD */
-        R_DX=((R_AX & 0x8000)?0xFFFF:0x0000);
-        NEXT;
-
-    _66_0x9C:                              /* PUSHFW */
-        CHECK_FLAGS(emu);
-        Push16(emu, (uint16_t)emu->eflags.x32);
-        NEXT;
-    _66_0x9D:                              /* POPFW */
-        CHECK_FLAGS(emu);
-        emu->eflags.x32 &=0xffff0000;
-        emu->eflags.x32 |= (Pop16(emu) & 0x3F7FD7) | 0x2;
-        NEXT;
-
-    _66_0xA1:                              /* MOV AX,Ow */
-        R_AX = *(uint16_t*)F32;
-        NEXT;
-    _66_0xA3:                              /* MOV Ow,AX */
-        *(uint16_t*)F32 = R_AX;
-        NEXT;
-    _66_0xA5:                              /* MOVSW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        *(uint16_t*)R_EDI = *(uint16_t*)R_ESI;
-        R_EDI += tmp8s;
-        R_ESI += tmp8s;
-        NEXT;
-
-    _66_0xA7:                              /* CMPSW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        tmp16u  = *(uint16_t*)R_EDI;
-        tmp16u2 = *(uint16_t*)R_ESI;
-        R_EDI += tmp8s;
-        R_ESI += tmp8s;
-        cmp16(emu, tmp16u2, tmp16u);
-        NEXT;
-
-    _66_0xA9:                             /* TEST AX,Iw */
-        test16(emu, R_AX, F16);
-        NEXT;
-
-    _66_0xAB:                              /* STOSW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        *(uint16_t*)R_EDI = R_AX;
-        R_EDI += tmp8s;
-        NEXT;
-    _66_0xAD:                              /* LODSW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        R_AX = *(uint16_t*)R_ESI;
-        R_ESI += tmp8s;
-        NEXT;
-    _66_0xAF:                              /* SCASW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        cmp16(emu, R_AX, *(uint16_t*)R_EDI);
-        R_EDI += tmp8s;
-        NEXT;
-
-    _66_0xB8:                              /* MOV AX,Iw */
-    _66_0xB9:                              /* MOV CX,Iw */
-    _66_0xBA:                              /* MOV DX,Iw */
-    _66_0xBB:                              /* MOV BX,Iw */
-    _66_0xBC:                              /*    ...     */
-    _66_0xBD:
-    _66_0xBE:
-    _66_0xBF:
-        emu->regs[opcode&7].word[0] = F16;
-        NEXT;
-
-    _66_0xC1:                              /* GRP2 Ew,Ib */
-        nextop = F8;
-        GET_EW;
-        tmp8u = F8 /*& 0x1f*/;
-        switch((nextop>>3)&7) {
-            case 0: EW->word[0] = rol16(emu, EW->word[0], tmp8u); break;
-            case 1: EW->word[0] = ror16(emu, EW->word[0], tmp8u); break;
-            case 2: EW->word[0] = rcl16(emu, EW->word[0], tmp8u); break;
-            case 3: EW->word[0] = rcr16(emu, EW->word[0], tmp8u); break;
-            case 4:
-            case 6: EW->word[0] = shl16(emu, EW->word[0], tmp8u); break;
-            case 5: EW->word[0] = shr16(emu, EW->word[0], tmp8u); break;
-            case 7: EW->word[0] = sar16(emu, EW->word[0], tmp8u); break;
-        }
-        NEXT;
-
-    _66_0xC7:                              /* MOV Ew,Iw */
-        nextop = F8;
-        GET_EW;
-        EW->word[0] = F16;
-        NEXT;
-
-    _66_0xCB:                               /* FAR RET */
-        ip = Pop(emu);
-        emu->segs[_CS] = Pop(emu);    // no check, no use....
-        emu->segs_serial[_CS] = 0;
-        // need to check status of CS register!
-        STEP
-        NEXT;
-    _66_0xCC:                              /* INT3 */
-        emu->old_ip = R_EIP;
-        R_EIP = ip-1;
-        if(my_context->signals[SIGTRAP])
-            raise(SIGTRAP);
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-    _66_0xD1:                              /* GRP2 Ew,1  */
-    _66_0xD3:                              /* GRP2 Ew,CL */
-        nextop = F8;
-        GET_EW;
-        tmp8u=(opcode==0xD3)?R_CL:1;
-        switch((nextop>>3)&7) {
-            case 0: EW->word[0] = rol16(emu, EW->word[0], tmp8u); break;
-            case 1: EW->word[0] = ror16(emu, EW->word[0], tmp8u); break;
-            case 2: EW->word[0] = rcl16(emu, EW->word[0], tmp8u); break;
-            case 3: EW->word[0] = rcr16(emu, EW->word[0], tmp8u); break;
-            case 4: 
-            case 6: EW->word[0] = shl16(emu, EW->word[0], tmp8u); break;
-            case 5: EW->word[0] = shr16(emu, EW->word[0], tmp8u); break;
-            case 7: EW->word[0] = sar16(emu, EW->word[0], tmp8u); break;
-        }
-        NEXT;
-    
-    _66_0xD9:
-        emu->old_ip = R_EIP;
-        R_EIP = ip;
-        Run66D9(emu);
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-
-    _66_0xDD:
-        emu->old_ip = R_EIP;
-        R_EIP = ip;
-        Run66DD(emu);
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-
-    _66_0xF0:                      /* LOCK prefix */
-        emu->old_ip = R_EIP;
-        R_EIP = ip - 2;
-        RunLock66(emu);
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-
-    _66_0xF2:                      /* REPNZ prefix */
-    _66_0xF3:                      /* REPZ prefix */
-        nextop = F8;
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        tmp32u = R_ECX;
-        switch(nextop) {
-            case 0xA4:              /* REP MOVSB */
-                while(tmp32u) {
-                    --tmp32u;
-                    *(uint8_t*)R_EDI = *(uint8_t*)R_ESI;
-                    R_EDI += tmp8s;
-                    R_ESI += tmp8s;
-                }
-                break;
-            case 0xA5:              /* REP MOVSW */
-                while(tmp32u) {
-                    --tmp32u;
-                    *(uint16_t*)R_EDI = *(uint16_t*)R_ESI;
-                    R_EDI += tmp8s;
-                    R_ESI += tmp8s;
-                }
-                break;
-            case 0xA7:              /* REP(N)Z CMPSW */
-                tmp16u = 0;
-                tmp16u2 = 0;
-                while(tmp32u) {
-                    --tmp32u;
-                    tmp16u  = *(uint16_t*)R_EDI;
-                    tmp16u2 = *(uint16_t*)R_ESI;
-                    R_EDI += tmp8s;
-                    R_ESI += tmp8s;
-                    if((tmp16u==tmp16u2)==(opcode==0xF2))
-                        break;
-                }
-                if(R_ECX) cmp16(emu, tmp16u2, tmp16u);
-                break;
-            case 0xAB:              /* REP STOSW */
-                while(tmp32u) {
-                    --tmp32u;
-                    *(uint16_t*)R_EDI = R_AX;
-                    R_EDI += tmp8s;
-                }
-                break;
-            case 0xAD:              /* REP LODSW */
-                while(tmp32u) {
-                    --tmp32u;
-                    R_AX = *(uint16_t*)R_ESI;
-                    R_ESI += tmp8s;
-                }
-                break;
-            case 0xAF:              /* REP(N)Z SCASW */
-                tmp16u = 0;
-                while(tmp32u) {
-                    --tmp32u;
-                    tmp16u = *(uint16_t*)R_EDI;
-                    R_EDI += tmp8s;
-                    if((R_AX==tmp16u)==(opcode==0xF2))
-                        break;
-                }
-                if(R_ECX) cmp16(emu, R_AX, tmp16u);
-                break;
-            default:
-                goto _default;
-        }
-        R_ECX = tmp32u;
-        NEXT;
-
-    _66_0xF5:                      /* CMC */
-        CHECK_FLAGS(emu);
-        CONDITIONAL_SET_FLAG(!ACCESS_FLAG(F_CF), F_CF);
-        NEXT;
-
-    _66_0xF7:                      /* GRP3 Ew(,Iw) */
-        nextop = F8;
-        GET_EW;
-        switch((nextop>>3)&7) {
-            case 0: 
-            case 1:                 /* TEST Ew,Iw */
-                test16(emu, EW->word[0], F16);
-                break;
-            case 2:                 /* NOT Ew */
-                EW->word[0] = not16(emu, EW->word[0]);
-                break;
-            case 3:                 /* NEG Ew */
-                EW->word[0] = neg16(emu, EW->word[0]);
-                break;
-            case 4:                 /* MUL AX,Ew */
-                mul16(emu, EW->word[0]);
-                break;
-            case 5:                 /* IMUL AX,Ew */
-                imul16_eax(emu, EW->word[0]);
-                break;
-            case 6:                 /* DIV Ew */
-                div16(emu, EW->word[0]);
-                break;
-            case 7:                 /* IDIV Ew */
-                idiv16(emu, EW->word[0]);
-                break;
-        }
-        NEXT;
-    _66_0xF8:                       /* CLC */
-        CHECK_FLAGS(emu);
-        CLEAR_FLAG(F_CF);
-        NEXT;
-    _66_0xF9:                       /* STC */
-        CHECK_FLAGS(emu);
-        SET_FLAG(F_CF);
-        NEXT;
-
-    _66_0xFF:                      /* GRP 5 Ew */
-        nextop = F8;
-        GET_EW;
-        switch((nextop>>3)&7) {
-            case 0:                 /* INC Ed */
-                EW->word[0] = inc16(emu, EW->word[0]);
-                break;
-            case 1:                 /* DEC Ed */
-                EW->word[0] = dec16(emu, EW->word[0]);
-                break;
-            case 6:
-                Push16(emu, EW->word[0]);
-                break;
-            default:
-                emu->old_ip = R_EIP;
-                R_EIP = ip;
-                printf_log(LOG_NONE, "Illegal Opcode 66 %02X %02X\n", opcode, nextop);
-                emu->quit=1;
-                emu->error |= ERR_ILLEGAL;
-                goto fini;
-        }
-        NEXT;

src/emu/x86emu.c

@@ -23,6 +23,13 @@
 #ifdef DYNAREC
 #include "custommem.h"
 #endif
+// for the applyFlushTo0
+#ifdef __i386__
+#include <immintrin.h>
+#elif defined(__arm__)
+#else
+#warning Architecture cannot follow SSE Flush to 0 flag
+#endif
 
 typedef struct cleanup_s {
     void*       f;
@@ -47,17 +54,6 @@ uint32_t* GetParityTab()
     return x86emu_parity_tab;
 }
 
-void PushExit(x86emu_t* emu)
-{
-    uintptr_t endMarker = AddCheckBridge(my_context->system, NULL, NULL, 0, "(ExitEmulation)");
-    Push(emu, endMarker);
-}
-
-void* GetExit()
-{
-    return (void*)AddCheckBridge(my_context->system, NULL, NULL, 0, "(ExitEmulation)");
-}
-
 static void internalX86Setup(x86emu_t* emu, box86context_t *context, uintptr_t start, uintptr_t stack, int stacksize, int ownstack)
 {
     emu->context = context;
@@ -81,13 +77,13 @@ static void internalX86Setup(x86emu_t* emu, box86context_t *context, uintptr_t s
     emu->segs[_GS] = 0x33;
     // setup fpu regs
     reset_fpu(emu);
-    emu->mxcsr = 0x1f80;  
+    emu->mxcsr.x32 = 0x1f80;  
 }
 
 EXPORTDYN
 x86emu_t *NewX86Emu(box86context_t *context, uintptr_t start, uintptr_t stack, int stacksize, int ownstack)
 {
-    printf_log(LOG_DEBUG, "Allocate a new X86 Emu, with EIP=%p and Stack=%p/0x%X\n", (void*)start, (void*)stack, stacksize);
+    printf_log(LOG_DEBUG, "%04d|Allocate a new X86 Emu, with EIP=%p and Stack=%p/0x%X\n", GetTID(), (void*)start, (void*)stack, stacksize);
 
     x86emu_t *emu = (x86emu_t*)box_calloc(1, sizeof(x86emu_t));
 
@@ -98,7 +94,7 @@ x86emu_t *NewX86Emu(box86context_t *context, uintptr_t start, uintptr_t stack, i
 
 x86emu_t *NewX86EmuFromStack(x86emu_t* emu, box86context_t *context, uintptr_t start, uintptr_t stack, int stacksize, int ownstack)
 {
-    printf_log(LOG_DEBUG, "New X86 Emu from stack, with EIP=%p and Stack=%p/0x%X\n", (void*)start, (void*)stack, stacksize);
+    printf_log(LOG_DEBUG, "%04d|New X86 Emu from stack, with EIP=%p and Stack=%p/0x%X\n", GetTID(), (void*)start, (void*)stack, stacksize);
 
     internalX86Setup(emu, context, start, stack, stacksize, ownstack);
     
@@ -108,6 +104,7 @@ x86emu_t *NewX86EmuFromStack(x86emu_t* emu, box86context_t *context, uintptr_t s
 EXPORTDYN
 void SetupX86Emu(x86emu_t *emu)
 {
+    (void)emu;
     printf_log(LOG_DEBUG, "Setup X86 Emu\n");
 }
 
@@ -128,6 +125,7 @@ void SetTraceEmu(uintptr_t start, uintptr_t end)
 
 void AddCleanup(x86emu_t *emu, void *p)
 {
+    (void)emu;
     if(my_context->clean_sz == my_context->clean_cap) {
         my_context->clean_cap += 4;
         my_context->cleanups = (cleanup_t*)box_realloc(my_context->cleanups, sizeof(cleanup_t)*my_context->clean_cap);
@@ -139,6 +137,7 @@ void AddCleanup(x86emu_t *emu, void *p)
 
 void AddCleanup1Arg(x86emu_t *emu, void *p, void* a)
 {
+    (void)emu;
     if(my_context->clean_sz == my_context->clean_cap) {
         my_context->clean_cap += 4;
         my_context->cleanups = (cleanup_t*)box_realloc(my_context->cleanups, sizeof(cleanup_t)*my_context->clean_cap);
@@ -188,6 +187,11 @@ void FreeX86Emu(x86emu_t **emu)
         return;
     printf_log(LOG_DEBUG, "%04d|Free a X86 Emu (%p)\n", GetTID(), *emu);
 
+    if((*emu)->test.emu) {
+        internalFreeX86((*emu)->test.emu);
+        box_free((*emu)->test.emu);
+        (*emu)->test.emu = NULL;
+    }
     internalFreeX86(*emu);
 
     box_free(*emu);
@@ -221,6 +225,10 @@ void CloneEmu(x86emu_t *newemu, const x86emu_t* emu)
 	newemu->top = emu->top;
     newemu->fpu_stack = emu->fpu_stack;
     memcpy(newemu->xmm, emu->xmm, sizeof(emu->xmm));
+    newemu->df = emu->df;
+    newemu->op1 = emu->op1;
+    newemu->op2 = emu->op2;
+    newemu->res = emu->res;
     newemu->mxcsr = emu->mxcsr;
     newemu->quit = emu->quit;
     newemu->error = emu->error;
@@ -230,6 +238,34 @@ void CloneEmu(x86emu_t *newemu, const x86emu_t* emu)
     newemu->regs[_SP].dword[0] = emu->regs[_SP].dword[0] + (intptr_t)(newst - oldst);
 }
 
+void CopyEmu(x86emu_t *newemu, const x86emu_t* emu)
+{
+	memcpy(newemu->regs, emu->regs, sizeof(emu->regs));
+    memcpy(&newemu->ip, &emu->ip, sizeof(emu->ip));
+	memcpy(&newemu->eflags, &emu->eflags, sizeof(emu->eflags));
+    newemu->old_ip = emu->old_ip;
+    memcpy(newemu->segs, emu->segs, sizeof(emu->segs));
+    memcpy(newemu->segs_serial, emu->segs_serial, sizeof(emu->segs_serial));
+    memcpy(newemu->segs_offs, emu->segs_offs, sizeof(emu->segs_offs));
+	memcpy(newemu->x87, emu->x87, sizeof(emu->x87));
+	memcpy(newemu->mmx, emu->mmx, sizeof(emu->mmx));
+    memcpy(newemu->xmm, emu->xmm, sizeof(emu->xmm));
+    memcpy(newemu->fpu_ld, emu->fpu_ld, sizeof(emu->fpu_ld));
+    memcpy(newemu->fpu_ll, emu->fpu_ll, sizeof(emu->fpu_ll));
+	memcpy(newemu->p_regs, emu->p_regs, sizeof(emu->p_regs));
+	newemu->cw = emu->cw;
+	newemu->sw = emu->sw;
+	newemu->top = emu->top;
+    newemu->fpu_stack = emu->fpu_stack;
+    newemu->df = emu->df;
+    newemu->op1 = emu->op1;
+    newemu->op2 = emu->op2;
+    newemu->res = emu->res;
+    newemu->mxcsr = emu->mxcsr;
+    newemu->quit = emu->quit;
+    newemu->error = emu->error;
+}
+
 box86context_t* GetEmuContext(x86emu_t* emu)
 {
     return emu->context;
@@ -301,7 +337,7 @@ void ResetFlags(x86emu_t *emu)
 const char* DumpCPURegs(x86emu_t* emu, uintptr_t ip)
 {
     static char buff[800];
-    char* regname[] = {"EAX", "ECX", "EDX", "EBX", "ESP", "EBP", "ESI", "EDI"};
+    static const char* regname[] = {"EAX", "ECX", "EDX", "EBX", "ESP", "EBP", "ESI", "EDI"};
     char tmp[80];
     buff[0] = '\0';
     if(trace_emm) {
@@ -458,6 +494,7 @@ static inline void init_perfcounters (int32_t do_reset, int32_t enable_divider)
 
 uint64_t ReadTSC(x86emu_t* emu)
 {
+    (void)emu;
     // Read the TimeStamp Counter as 64bits.
     // this is supposed to be the number of instructions executed since last reset
 #if defined(__i386__)
@@ -497,3 +534,17 @@ void ResetSegmentsCache(x86emu_t *emu)
         return;
     memset(emu->segs_serial, 0, sizeof(emu->segs_serial));
 }
+
+void applyFlushTo0(x86emu_t* emu)
+{
+#ifdef __i386__
+    _mm_setcsr(_mm_getcsr() | (emu->mxcsr.x32&0x8040));
+#elif defined(__arm__)
+    uint64_t fpscr = __builtin_arm_get_fpscr();
+    fpscr &= ~(1<<24);                     // clear bit FZ (24)
+    fpscr |= (emu->mxcsr.f.MXCSR_FZ)<<24;  // set FZ as mxcsr FZ
+    __builtin_arm_set_fpscr(fpscr);
+#else
+    (void)emu;
+#endif
+}

src/emu/x86emu_private.h

@@ -23,6 +23,17 @@ typedef struct forkpty_s {
     void*   f;  // forkpty function
 } forkpty_t;
 
+typedef struct x86emu_s x86emu_t;
+
+typedef struct x86test_s {
+    x86emu_t*   emu;
+    uintptr_t   memaddr;
+    int         memsize;
+    int         test;
+    int         clean;
+    uint8_t     mem[16];
+} x86test_t;
+
 typedef struct x86emu_s {
     // cpu
 	reg32_t     regs[8];
@@ -40,7 +51,7 @@ typedef struct x86emu_s {
 	fpu_p_reg_t p_regs[8];
     // sse
     sse_regs_t  xmm[8];
-    uint32_t    mxcsr;
+    mmxcontrol_t mxcsr;
     uintptr_t   old_ip;
     // defered flags
     defered_flags_t df;
@@ -63,6 +74,7 @@ typedef struct x86emu_s {
     int         exit;
     int         quitonlongjmp;  // quit if longjmp is called
     int         longjmp;        // if quit because of longjmp
+    x86test_t   test;       // used for dynarec testing
     // parent context
     box86context_t *context;
     // cpu helpers
@@ -88,4 +100,6 @@ typedef struct x86emu_s {
 //#define INTR_RAISE_DIV0(emu) {emu->error |= ERR_DIVBY0; emu->quit=1;}
 #define INTR_RAISE_DIV0(emu) {emu->error |= ERR_DIVBY0;} // should rise a SIGFPE and not quit
 
+void applyFlushTo0(x86emu_t* emu);
+
 #endif //__X86EMU_PRIVATE_H_

src/emu/x86int3.c

@@ -10,6 +10,7 @@
 #include <sys/types.h>
 #include <pthread.h>
 #include <signal.h>
+#include <wait.h>
 
 #include "debug.h"
 #include "box86stack.h"
@@ -53,7 +54,10 @@ x86emu_t* x86emu_fork(x86emu_t* emu, int forktype)
         for (int i=0; i<my_context->atfork_sz; --i)
             if(my_context->atforks[i].parent)
                 EmuCall(emu, my_context->atforks[i].parent);
-
+        if(forktype==3) {
+            // vfork, the parent wait the end or execve of the son
+            waitpid(v, NULL, WEXITED);
+        }
     } else if(v==0) {
         // execute atforks child functions
         for (int i=0; i<my_context->atfork_sz; --i)
@@ -87,6 +91,8 @@ void x86Int3(x86emu_t* emu)
             if(h && strstr(ElfName(h), "libMiles")) have_trace = 1;*/
             if(box86_log>=LOG_DEBUG  || cycle_log) {
                 int tid = GetTID();
+                const char *s = NULL;
+                s = GetNativeName((void*)addr);
                 char t_buff[256] = "\0";
                 char buff2[64] = "\0";
                 char buff3[64] = "\0";
@@ -99,210 +105,215 @@ void x86Int3(x86emu_t* emu)
                 int post = 0;
                 int perr = 0;
                 uint32_t *pu32 = NULL;
-                const char *s = NULL;
-                s = GetNativeName((void*)addr);
                 if(addr==(uintptr_t)PltResolver) {
                     snprintf(buff, 256, "%s", cycle_log?"PltResolver ":" ... ");
                 } else
                 if(strstr(s, "SDL_RWFromFile")==s || strstr(s, "SDL_RWFromFile")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%s, %s)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%s, %s)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else  if(strstr(s, "glColor4f")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f, %f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8), *(float*)(R_ESP+12), *(float*)(R_ESP+16));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f, %f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8), *(float*)(R_ESP+12), *(float*)(R_ESP+16));
                 } else  if(strstr(s, "glTexCoord2f")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
                 } else  if(strstr(s, "glVertex2f")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
                 } else  if(strstr(s, "glVertex3f")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8), *(float*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8), *(float*)(R_ESP+12));
                 } else  if(strstr(s, "__open64")==s || strcmp(s, "open64")==0) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(int*)(R_ESP+8), *(int*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(int*)(R_ESP+8), *(int*)(R_ESP+12));
                     perr = 1;
                 } else  if(!strcmp(s, "opendir")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                     perr = 1;
                 } else  if(strstr(s, "__open")==s || !strcmp(s, "open") || !strcmp(s, "my_open64")) {
                     tmp = *(char**)(R_ESP+4);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d (,%d))", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8), *(int*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %d (,%d))", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8), *(int*)(R_ESP+12));
                     perr = 1;
                 } else  if(!strcmp(s, "shm_open")) {
                     tmp = *(char**)(R_ESP+4);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8), *(int*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8), *(int*)(R_ESP+12));
                     perr = 1;
                 } else  if(strcmp(s, "mkdir")==0) {
                     tmp = *(char**)(R_ESP+4);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d)", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %d)", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8));
                     perr = 1;
                 } else  if(!strcmp(s, "fopen")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                     perr = 2;
                 } else  if(!strcmp(s, "freopen")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\", %p)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\", %p)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                     perr = 2;
                 } else  if(!strcmp(s, "fopen64")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                     perr = 2;
                 } else  if(!strcmp(s, "chdir")) {
                     pu32=*(uint32_t**)(R_ESP+4);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, pu32?((pu32==(uint32_t*)1)?"/1/":(char*)pu32):"/0/");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, pu32?((pu32==(uint32_t*)1)?"/1/":(char*)pu32):"/0/");
                 } else  if(strstr(s, "getenv")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                     post = 2;
                 } else  if(strstr(s, "putenv")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                 } else  if(strstr(s, "pread")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p, %u, %d)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12), *(int32_t*)(R_ESP+16));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p, %u, %d)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12), *(int32_t*)(R_ESP+16));
                     perr = 1;
                 } else  if(!strcmp(s, "read")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p, %u)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p, %u)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12));
                     perr = 1;
                 } else  if(strstr(s, "ioctl")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, 0x%x, %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(int32_t*)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, 0x%x, %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(int32_t*)(R_ESP+8), *(void**)(R_ESP+12));
                     perr = 1;
                 } else  if(strstr(s, "statvfs64")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+4), *(void**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+4), *(void**)(R_ESP+8));
                 } else  if(strstr(s, "index")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p(\"%s\"), %i(%c))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+4), *(int32_t*)(R_ESP+8), *(int32_t*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p(\"%s\"), %i(%c))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+4), *(int32_t*)(R_ESP+8), *(int32_t*)(R_ESP+8));
                 } else  if(strstr(s, "rindex")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p(\"%s\"), %i(%c))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+4), *(int32_t*)(R_ESP+8), *(int32_t*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p(\"%s\"), %i(%c))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+4), *(int32_t*)(R_ESP+8), *(int32_t*)(R_ESP+8));
                 } else  if(strstr(s, "__xstat64")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                     perr = 1;
                 } else  if(strcmp(s, "__xstat")==0) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                     perr = 1;
                 } else  if(strstr(s, "__lxstat64")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                     perr = 1;
                 } else  if(strstr(s, "sem_timedwait")==s) {
                     pu32 = *(uint32_t**)(R_ESP+8);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %p[%d sec %d ns])", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), pu32?pu32[0]:-1, pu32?pu32[1]:-1);
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %p[%d sec %d ns])", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), pu32?(int32_t)pu32[0]:-1, pu32?(int32_t)pu32[1]:-1);
                     perr = 1;
                 } else  if(strstr(s, "waitpid")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p, 0x%x)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p, 0x%x)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12));
                     perr = 1;
                 } else  if(strstr(s, "clock_gettime")==s || strstr(s, "__clock_gettime")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(void**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(void**)(R_ESP+8));
                     post = 1;
                     pu32 = *(uint32_t**)(R_ESP+8);
                 } else  if(strstr(s, "semop")==s) {
                     int16_t* p16 = *(int16_t**)(R_ESP+8);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p[%u/%d/0x%x], %d)", tid, *(void**)(R_ESP), s, *(int*)(R_ESP+4), p16, p16[0], p16[1], p16[2], *(int*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p[%u/%d/0x%x], %d)", tid, *(void**)(R_ESP), s, *(int*)(R_ESP+4), p16, p16[0], p16[1], p16[2], *(int*)(R_ESP+12));
                     perr = 1;
                 } else  if(!strcmp(s, "mmap64")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, 0x%x, %d, 0x%x, %d, %lld)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int64_t*)(R_ESP+24));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, 0x%x, %d, 0x%x, %d, %lld)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int64_t*)(R_ESP+24));
                     perr = 3;
                 } else  if(!strcmp(s, "mmap")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, 0x%x, %d, 0x%x, %d, %d)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int*)(R_ESP+24));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, 0x%x, %d, 0x%x, %d, %d)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int*)(R_ESP+24));
                     perr = 3;
                 } else  if(strstr(s, "strcasecmp")==s || strstr(s, "__strcasecmp")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else  if(strstr(s, "gtk_signal_connect_full")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\", %p, %p, %p, %p, %d, %d)", tid, *(void**)(R_ESP), "gtk_signal_connect_full", *(void**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16), *(void**)(R_ESP+20), *(void**)(R_ESP+24), *(int32_t*)(R_ESP+28), *(int32_t*)(R_ESP+32));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\", %p, %p, %p, %p, %d, %d)", tid, *(void**)(R_ESP), "gtk_signal_connect_full", *(void**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16), *(void**)(R_ESP+20), *(void**)(R_ESP+24), *(int32_t*)(R_ESP+28), *(int32_t*)(R_ESP+32));
                 } else  if(strstr(s, "strcmp")==s || strstr(s, "__strcmp")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else  if(strstr(s, "strstr")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%.127s\", \"%.127s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%.127s\", \"%.127s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else  if(strstr(s, "strlen")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p(\"%s\"))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), ((R_ESP+4))?(*(char**)(R_ESP+4)):"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p(\"%s\"))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), ((R_ESP+4))?(*(char**)(R_ESP+4)):"nil");
                 } else  if(strstr(s, "vsnprintf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%08X, %u, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8), *(uint32_t*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%08X, %u, %p(\"%s\")...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8), *(void**)(R_ESP+12), *(char**)(R_ESP+12));
                     pu32 = *(uint32_t**)(R_ESP+4);
                     post = 3;
                 } else  if(strstr(s, "vsprintf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                     pu32 = *(uint32_t**)(R_ESP+4);
                     post = 3;
                 } else  if(strstr(s, "__vsprintf_chk")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %d, %zu, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(int*)(R_ESP+8), *(size_t*)(R_ESP+12), *(char**)(R_ESP+16), *(void**)(R_ESP+20));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %d, %zu, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(int*)(R_ESP+8), *(size_t*)(R_ESP+12), *(char**)(R_ESP+16), *(void**)(R_ESP+20));
                     pu32 = *(uint32_t**)(R_ESP+4);
                     post = 3;
                 } else  if(strstr(s, "__snprintf_chk")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, %d, %d, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(char**)(R_ESP+20), *(void**)(R_ESP+24));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %zu, %d, %d, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(char**)(R_ESP+20), *(void**)(R_ESP+24));
                     pu32 = *(uint32_t**)(R_ESP+4);
                     post = 3;
                 } else  if(strstr(s, "snprintf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(char**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %zu, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(char**)(R_ESP+12));
                     pu32 = *(uint32_t**)(R_ESP+4);
                     post = 3;
                 } else  if(strstr(s, "sprintf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%08X, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%08X, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8));
                     pu32 = *(uint32_t**)(R_ESP+4);
                     post = 3;
                 } else  if(strstr(s, "printf")==s) {
                     pu32 = *(uint32_t**)(R_ESP+4);
                     if(((uintptr_t)pu32)<0x5) // probably a _chk function
                         pu32 = *(uint32_t**)(R_ESP+8);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\"...)", tid, *(void**)(R_ESP), s, pu32?((char*)(pu32)):"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\"...)", tid, *(void**)(R_ESP), s, pu32?((char*)(pu32)):"nil");
                 } else  if(strstr(s, "wprintf")==s) {
                     pu32 = *(uint32_t**)(R_ESP+4);
                     if(((uintptr_t)pu32)<0x5) // probably a _chk function
                         pu32 = *(uint32_t**)(R_ESP+8);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%S\"...)", tid, *(void**)(R_ESP), s, pu32?((wchar_t*)(pu32)):L"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%S\"...)", tid, *(void**)(R_ESP), s, pu32?((wchar_t*)(pu32)):L"nil");
                 } else  if(strstr(s, "__vswprintf")==s) {
                     if(*(size_t*)(R_ESP+12)<2) {
-                        snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, %p, %p, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16), *(void**)(R_ESP+20));
+                        snprintf(buff, 256, "%04d|%p: Calling %s(%p, %zu, %p, %p, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16), *(void**)(R_ESP+20));
                     } else {
-                        snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, \"%S\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(wchar_t**)(R_ESP+12), *(void**)(R_ESP+16));
+                        snprintf(buff, 256, "%04d|%p: Calling %s(%p, %zu, \"%S\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(wchar_t**)(R_ESP+12), *(void**)(R_ESP+16));
                         pu32 = *(uint32_t**)(R_ESP+4);
                         post = 6;
                     }
                 } else  if(strstr(s, "puts")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\"...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\"...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                 } else  if(strstr(s, "fputs")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %p...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(void**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %p...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(void**)(R_ESP+8));
                 } else  if(strstr(s, "fprintf")==s) {
                     pu32 = *(uint32_t**)(R_ESP+8);
                     if(((uintptr_t)pu32)<0x5) // probably a __fprint_chk
                         pu32 = *(uint32_t**)(R_ESP+12);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%08X, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), pu32?((char*)(pu32)):"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%08X, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), pu32?((char*)(pu32)):"nil");
                 } else  if(strstr(s, "vfprintf")==s) {
                     pu32 = *(uint32_t**)(R_ESP+8);
                     if(((uintptr_t)pu32)<0x5) // probably a _chk function
                         pu32 = *(uint32_t**)(R_ESP+12);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%08X, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), pu32?((char*)(pu32)):"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%08X, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), pu32?((char*)(pu32)):"nil");
                 } else  if(strstr(s, "vkGetInstanceProcAddr")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else  if(strstr(s, "vkGetDeviceProcAddr")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else  if(strstr(s, "glXGetProcAddress")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                 } else  if(strstr(s, "sscanf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\", ...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\", ...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else  if(!strcmp(s, "vsscanf")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\", ...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\", ...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else if(strstr(s, "XCreateWindow")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %p, %d, %d, %u, %u, %u, %d, %u, %p, %u, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(uint32_t*)(R_ESP+20), *(uint32_t*)(R_ESP+24), *(uint32_t*)(R_ESP+28), *(int32_t*)(R_ESP+32), *(uint32_t*)(R_ESP+36), *(void**)(R_ESP+40), *(uint32_t*)(R_ESP+44), *(void**)(R_ESP+48));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %p, %d, %d, %u, %u, %u, %d, %u, %p, %u, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(uint32_t*)(R_ESP+20), *(uint32_t*)(R_ESP+24), *(uint32_t*)(R_ESP+28), *(int32_t*)(R_ESP+32), *(uint32_t*)(R_ESP+36), *(void**)(R_ESP+40), *(uint32_t*)(R_ESP+44), *(void**)(R_ESP+48));
                 } else if(strstr(s, "XLoadQueryFont")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else if(strstr(s, "pthread_mutex_lock")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4));
                 } else if(!strcmp(s, "fmodf")) {
                     post = 4;
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
                 } else if(!strcmp(s, "fmod")) {
                     post = 4;
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(double*)(R_ESP+4), *(double*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(double*)(R_ESP+4), *(double*)(R_ESP+12));
                 } else if(strstr(s, "SDL_GetWindowSurface")==s) {
                     post = 5;
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4));
                 } else if(strstr(s, "udev_monitor_new_from_netlink")==s) {
                     post = 5;
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
                 } else  if(!strcmp(s, "syscall")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p, %p, %p...)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p, %p, %p...)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16));
                     perr = 1;
+                } else  if(strstr(s, "___tls_get_addr")) {
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p[%d, %d])", tid, *(void**)(R_ESP), s, (void*)(R_EAX), ((int*)(R_EAX))[0], ((int*)(R_EAX))[1]);
+                } else  if(strstr(s, "__tls_get_addr")) {
+                    pu32 = *(uint32_t**)(R_ESP+4);
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p[%d, %d])", tid, *(void**)(R_ESP), s, pu32, pu32[0], pu32[1]);
+                } else if (!strcmp(s, "XDrawImageString")) {
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %p, %p, %i, %i, %p\"%s\", %d)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), *(void**)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(char**)(R_ESP+24), *(char**)(R_ESP+24), *(int*)(R_ESP+28));
                 } else {
-                    snprintf(buff, 255, "%04d|%p: Calling %s (%08X, %08X, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8), *(uint32_t*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s (%08X, %08X, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8), *(uint32_t*)(R_ESP+12));
                 }
                 if(!cycle_log) {
-                    pthread_mutex_lock(&emu->context->mutex_trace);
+                    mutex_lock(&emu->context->mutex_trace);
                     printf_log(LOG_NONE, "%s =>", buff);
-                    pthread_mutex_unlock(&emu->context->mutex_trace);
+                    mutex_unlock(&emu->context->mutex_trace);
                 }
                 w(emu, addr);   // some function never come back, so unlock the mutex first!
                 if(post)
                     switch(post) {
-                    case 1: snprintf(buff2, 63, " [%d sec %d nsec]", pu32?pu32[0]:-1, pu32?pu32[1]:-1);
+                    case 1: snprintf(buff2, 63, " [%d sec %d nsec]", pu32?(int32_t)pu32[0]:-1, pu32?(int32_t)pu32[1]:-1);
                             break;
                     case 2: snprintf(buff2, 63, "(%s)", R_EAX?((char*)R_EAX):"nil");
                             break;
@@ -330,9 +341,9 @@ void x86Int3(x86emu_t* emu)
                 if(cycle_log)
                     snprintf(buffret, 127, "0x%X%s%s", R_EAX, buff2, buff3);
                 else {
-                    pthread_mutex_lock(&emu->context->mutex_trace);
+                    mutex_lock(&emu->context->mutex_trace);
                     printf_log(LOG_NONE, " return 0x%08X%s%s\n", R_EAX, buff2, buff3);
-                    pthread_mutex_unlock(&emu->context->mutex_trace);
+                    mutex_unlock(&emu->context->mutex_trace);
                 }
             } else
                 w(emu, addr);
@@ -362,4 +373,4 @@ void print_cycle_log(int loglevel) {
             }
         }
     }
-}
+}

src/emu/x86primop.c

@@ -1044,14 +1044,14 @@ Implements the SBB instruction and side effects.
 ****************************************************************************/
 uint16_t sbb16(x86emu_t *emu, uint16_t d, uint16_t s)
 {
-    uint32_t res;   /* all operands in native machine order */
-    uint32_t bc;
+	uint32_t res;   /* all operands in native machine order */
+	uint32_t bc;
 	CHECK_FLAGS(emu);
 
 	if (ACCESS_FLAG(F_CF))
-        res = d - s - 1;
-    else
-        res = d - s;
+		res = d - s - 1;
+	else
+		res = d - s;
 	CONDITIONAL_SET_FLAG(res & 0x8000, F_SF);
 	CONDITIONAL_SET_FLAG((res & 0xffff) == 0, F_ZF);
 	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
@@ -1075,9 +1075,9 @@ uint32_t sbb32(x86emu_t *emu, uint32_t d, uint32_t s)
 	CHECK_FLAGS(emu);
 
 	if (ACCESS_FLAG(F_CF))
-        res = d - s - 1;
-    else
-        res = d - s;
+		res = d - s - 1;
+	else
+		res = d - s;
 	CONDITIONAL_SET_FLAG(res & 0x80000000, F_SF);
 	CONDITIONAL_SET_FLAG(!res, F_ZF);
 	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);

src/emu/x86primop.h

@@ -247,6 +247,7 @@ Implements the NOT instruction
 ****************************************************************************/
 static inline uint8_t not8(x86emu_t *emu, uint8_t s)
 {
+    (void)emu;
 	return ~s;
 }
 
@@ -256,6 +257,7 @@ Implements the NOT instruction
 ****************************************************************************/
 static inline uint16_t not16(x86emu_t *emu, uint16_t s)
 {
+    (void)emu;
 	return ~s;
 }
 
@@ -265,6 +267,7 @@ Implements the NOT instruction
 ****************************************************************************/
 static inline uint32_t not32(x86emu_t *emu, uint32_t s)
 {
+    (void)emu;
 	return ~s;
 }
 

src/emu/x86run64.c

@@ -0,0 +1,459 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+#include "bridge.h"
+#ifdef DYNAREC
+#include "../dynarec/arm_lock_helper.h"
+#endif
+
+#include "modrm.h"
+
+
+#ifdef TEST_INTERPRETER
+uintptr_t Test64(x86test_t *test, int seg, uintptr_t addr)
+#else
+uintptr_t Run64(x86emu_t *emu, int seg, uintptr_t addr)
+#endif
+{
+    uint8_t opcode = F8;
+    uint8_t nextop;
+    reg32_t *oped;
+    uint8_t tmp8u;
+    uint32_t tmp32u;
+    int32_t tmp32s;
+    #ifdef TEST_INTERPRETER
+    x86emu_t* emu = test->emu;
+    #endif
+    uintptr_t tlsdata = (seg==_FS)?GetFSBaseEmu(emu):GetGSBaseEmu(emu);
+    switch(opcode) {
+        case 0x01:              /* ADD GS:Ed, Gd */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = add32(emu, ED->dword[0], GD.dword[0]);
+            break;
+
+        case 0x03:              /* ADD Gd, GS:Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            GD.dword[0] = add32(emu, GD.dword[0], ED->dword[0]);
+            break;
+
+        case 0x06:              /* PUSH ES */
+            Push(emu, emu->segs[_ES]);
+            break;
+        case 0x07:             /* POP ES */
+            emu->segs[_ES] = Pop(emu);
+            emu->segs_serial[_ES] = 0;
+            break;
+
+        case 0x0f:
+            #ifdef TEST_INTERPRETER
+            return Test640F(test, tlsdata, addr);
+            #else
+            return Run640F(emu, tlsdata, addr);
+            #endif
+
+        case 0x11:              /* ADC GS:Ed, Gd */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = adc32(emu, ED->dword[0], GD.dword[0]);
+            break;
+        
+        case 0x13:              /* ADC Gd, GS:Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            GD.dword[0] = adc32(emu, GD.dword[0], ED->dword[0]);
+            break;
+
+        case 0x21:              /* AND Ed,Gd */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = and32(emu, ED->dword[0], GD.dword[0]);
+            break;
+
+        case 0x23:              /* AND Gd,Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            GD.dword[0] = and32(emu, GD.dword[0], ED->dword[0]);
+            break;
+
+        case 0x29:              /* SUB GS:Ed, Gd */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = sub32(emu, ED->dword[0], GD.dword[0]);
+            break;
+
+        case 0x2B:              /* SUB Gd, GS:Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            GD.dword[0] = sub32(emu, GD.dword[0], ED->dword[0]);
+            break;
+
+        case 0x31:              /* XOR Ed,Gd */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = xor32(emu, ED->dword[0], GD.dword[0]);
+            break;
+
+        case 0x33:              /* XOR Gd,Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            GD.dword[0] = xor32(emu, GD.dword[0], ED->dword[0]);
+            break;
+
+        case 0x39:              /* CMP Ed,Gd */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            cmp32(emu, ED->dword[0], GD.dword[0]);
+            break;
+
+        case 0x3B:              /* CMP Gd,Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            cmp32(emu, GD.dword[0], ED->dword[0]);
+            break;
+        
+        case 0x40:
+        case 0x41:
+        case 0x42:
+        case 0x43:
+        case 0x44:
+        case 0x45:
+        case 0x46:
+        case 0x47:             /* INC Reg */
+            tmp8u = opcode&7;
+            emu->regs[tmp8u].dword[0] = inc32(emu, emu->regs[tmp8u].dword[0]);
+            break;
+        case 0x48:
+        case 0x49:
+        case 0x4A:
+        case 0x4B:
+        case 0x4C:
+        case 0x4D:
+        case 0x4E:
+        case 0x4F:            /* DEC Reg */
+            tmp8u = opcode&7;
+            emu->regs[tmp8u].dword[0] = dec32(emu, emu->regs[tmp8u].dword[0]);
+            break;
+        case 0x50:              /* PUSH Reg */
+        case 0x51:
+        case 0x52:
+        case 0x53:
+        case 0x54:               /* PUSH ESP */
+        case 0x55:
+        case 0x56:
+        case 0x57:              
+        case 0x58:              /* POP Reg */
+        case 0x59:
+        case 0x5A:
+        case 0x5B:
+        case 0x5C:              /* POP ESP */
+        case 0x5D:
+        case 0x5E:
+        case 0x5F:
+            // Segment override if for memory loc, no stack segment 
+            return addr-1;
+
+        case 0x64:              /* FS: */
+            // so just ignore that GS: prefix then
+            #ifdef TEST_INTERPRETER
+            return Test64(test, _FS, addr-1); // put FS back
+            #else
+            return Run64(emu, _FS, addr-1); // put FS back
+            #endif
+
+        case 0x66:
+            #ifdef TEST_INTERPRETER
+            return Test6466(test, tlsdata, addr);
+            #else
+            return Run6466(emu, tlsdata, addr);
+            #endif
+        case 0x67:
+            #ifdef TEST_INTERPRETER
+            return Test6467(test, tlsdata, addr);
+            #else
+            return Run6467(emu, tlsdata, addr);
+            #endif
+
+        case 0x69:              /* IMUL Gd,Ed,Id */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            tmp32u = F32;
+            GD.dword[0] = imul32(emu, ED->dword[0], tmp32u);
+            break;
+        
+        case 0x70:
+        case 0x71:
+        case 0x72:
+        case 0x73:
+        case 0x74:
+        case 0x75:
+        case 0x76:
+        case 0x77:
+        case 0x78:
+        case 0x79:
+        case 0x7A:
+        case 0x7B:
+        case 0x7C:
+        case 0x7D:
+        case 0x7E:
+        case 0x7F:
+            // just ignore the prefix
+            return addr-1;
+        case 0x80:             /* GRP Eb,Ib */
+        case 0x82:             // 0x82 and 0x80 are the same opcodes it seems?
+            nextop = F8;
+            GET_EB_OFFS(tlsdata);
+            tmp8u = F8;
+            switch((nextop>>3)&7) {
+                case 0: EB->byte[0] = add8(emu, EB->byte[0], tmp8u); break;
+                case 1: EB->byte[0] =  or8(emu, EB->byte[0], tmp8u); break;
+                case 2: EB->byte[0] = adc8(emu, EB->byte[0], tmp8u); break;
+                case 3: EB->byte[0] = sbb8(emu, EB->byte[0], tmp8u); break;
+                case 4: EB->byte[0] = and8(emu, EB->byte[0], tmp8u); break;
+                case 5: EB->byte[0] = sub8(emu, EB->byte[0], tmp8u); break;
+                case 6: EB->byte[0] = xor8(emu, EB->byte[0], tmp8u); break;
+                case 7:               cmp8(emu, EB->byte[0], tmp8u); break;
+            }
+            break;
+        case 0x81:              /* GRP Ed,Id */
+        case 0x83:              /* GRP Ed,Ib */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            if(opcode==0x83) {
+                tmp32s = F8S;
+                tmp32u = (uint32_t)tmp32s;
+            } else
+                tmp32u = F32;
+            switch((nextop>>3)&7) {
+                case 0: ED->dword[0] = add32(emu, ED->dword[0], tmp32u); break;
+                case 1: ED->dword[0] =  or32(emu, ED->dword[0], tmp32u); break;
+                case 2: ED->dword[0] = adc32(emu, ED->dword[0], tmp32u); break;
+                case 3: ED->dword[0] = sbb32(emu, ED->dword[0], tmp32u); break;
+                case 4: ED->dword[0] = and32(emu, ED->dword[0], tmp32u); break;
+                case 5: ED->dword[0] = sub32(emu, ED->dword[0], tmp32u); break;
+                case 6: ED->dword[0] = xor32(emu, ED->dword[0], tmp32u); break;
+                case 7:                cmp32(emu, ED->dword[0], tmp32u); break;
+            }
+            break;
+        case 0x87:
+            nextop = F8;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+            GET_ED_OFFS(tlsdata);
+            if((nextop&0xC0)==0xC0) {
+                tmp32u = GD.dword[0];
+                GD.dword[0] = ED->dword[0];
+                ED->dword[0] = tmp32u;
+            } else {
+                if(((uintptr_t)ED)&3)
+                {
+                    // not aligned, dont't try to "LOCK"
+                    tmp32u = ED->dword[0];
+                    ED->dword[0] = GD.dword[0];
+                    GD.dword[0] = tmp32u;
+                } else {
+                    // XCHG is supposed to automaticaly LOCK memory bus
+                    GD.dword[0] = arm_lock_xchg(ED, GD.dword[0]);
+                }
+            }
+#else
+            GET_ED_OFFS(tlsdata);
+            if((nextop&0xC0)!=0xC0)
+                pthread_mutex_lock(&emu->context->mutex_lock); // XCHG always LOCK (but when accessing memory only)
+            tmp32u = GD.dword[0];
+            GD.dword[0] = ED->dword[0];
+            ED->dword[0] = tmp32u;
+            if((nextop&0xC0)!=0xC0)
+                pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+            break;
+            
+        case 0x89:              /* MOV Ed,Gd */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = GD.dword[0];
+            break;
+        case 0x8A:              /* MOV Gb,GS:Eb */
+            nextop = F8;
+            GET_EB_OFFS(tlsdata);
+            GB = EB->byte[0];
+            break;
+        case 0x8B:              /* MOV Gd,Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            GD.dword[0] = ED->dword[0];
+            break;
+        case 0x8C:              /* MOV Seg:Ed, Seg */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = emu->segs[((nextop&0x38)>>3)];
+            break;
+        case 0x8E:              /* MOV Seg,Seg:Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            emu->segs[((nextop&0x38)>>3)] = ED->word[0];
+            emu->segs_serial[((nextop&0x38)>>3)] = 0;
+            break;
+        case 0x8F:              /* POP Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = Pop(emu);
+            break;
+
+        case 0x9C:              /* PUSHFD */
+            // Segment override if for memory loc, no stack segment 
+            return addr-1;   // so ignore prefix and continue
+
+        case 0xA0:              /* MOV AL,Ob */
+            tmp32s = F32S;
+            R_AL = *(uint8_t*)((tlsdata) + tmp32s);
+            break;
+        case 0xA1:              /* MOV EAX,Ov */
+            tmp32s = F32S;
+            R_EAX = *(uint32_t*)((tlsdata) + tmp32s);
+            break;
+        case 0xA2:              /* MOV Ob,AL */
+            tmp32s = F32S;
+            *(uint8_t*)((tlsdata) + tmp32s) = R_AL;
+            break;
+        case 0xA3:             /* MOV Od,EAX */
+            tmp32s = F32S;
+            *(uint32_t*)((tlsdata) + tmp32s) = R_EAX;
+            break;
+
+        case 0xA8:            /* TEST AL, Ib */
+            test8(emu, R_AL, F8);
+            break;
+
+        case 0xC0:             /* GRP2 Eb,Ib */
+            nextop = F8;
+            GET_EB_OFFS(tlsdata);
+            tmp8u = F8/* & 0x1f*/; // masking done in each functions
+            switch((nextop>>3)&7) {
+                case 0: EB->byte[0] = rol8(emu, EB->byte[0], tmp8u); break;
+                case 1: EB->byte[0] = ror8(emu, EB->byte[0], tmp8u); break;
+                case 2: EB->byte[0] = rcl8(emu, EB->byte[0], tmp8u); break;
+                case 3: EB->byte[0] = rcr8(emu, EB->byte[0], tmp8u); break;
+                case 4:
+                case 6: EB->byte[0] = shl8(emu, EB->byte[0], tmp8u); break;
+                case 5: EB->byte[0] = shr8(emu, EB->byte[0], tmp8u); break;
+                case 7: EB->byte[0] = sar8(emu, EB->byte[0], tmp8u); break;
+            }
+            break;
+
+        case 0xC6:              /* MOV Eb,Ib */
+            nextop = F8;
+            GET_EB_OFFS(tlsdata);
+            EB->byte[0] = F8;
+            break;
+        case 0xC7:              /* MOV Ed,Id */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            ED->dword[0] = F32;
+            break;
+
+        case 0xD0:                      /* GRP2 Eb,1 */
+        case 0xD2:                      /* GRP2 Eb,CL */
+            nextop = F8;
+            GET_EB_OFFS(tlsdata);
+            tmp8u = (opcode==0xD0)?1:R_CL;
+            switch((nextop>>3)&7) {
+                case 0: EB->byte[0] = rol8(emu, EB->byte[0], tmp8u); break;
+                case 1: EB->byte[0] = ror8(emu, EB->byte[0], tmp8u); break;
+                case 2: EB->byte[0] = rcl8(emu, EB->byte[0], tmp8u); break;
+                case 3: EB->byte[0] = rcr8(emu, EB->byte[0], tmp8u); break;
+                case 4: 
+                case 6: EB->byte[0] = shl8(emu, EB->byte[0], tmp8u); break;
+                case 5: EB->byte[0] = shr8(emu, EB->byte[0], tmp8u); break;
+                case 7: EB->byte[0] = sar8(emu, EB->byte[0], tmp8u); break;
+            }
+            break;
+        case 0xD1:                      /* GRP2 Ed,1 */
+        case 0xD3:                      /* GRP2 Ed,CL */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            tmp8u = (opcode==0xD1)?1:R_CL;
+            switch((nextop>>3)&7) {
+                case 0: ED->dword[0] = rol32(emu, ED->dword[0], tmp8u); break;
+                case 1: ED->dword[0] = ror32(emu, ED->dword[0], tmp8u); break;
+                case 2: ED->dword[0] = rcl32(emu, ED->dword[0], tmp8u); break;
+                case 3: ED->dword[0] = rcr32(emu, ED->dword[0], tmp8u); break;
+                case 4: 
+                case 6: ED->dword[0] = shl32(emu, ED->dword[0], tmp8u); break;
+                case 5: ED->dword[0] = shr32(emu, ED->dword[0], tmp8u); break;
+                case 7: ED->dword[0] = sar32(emu, ED->dword[0], tmp8u); break;
+            }
+            break;
+
+        case 0xE9:
+        case 0xEB:
+            return addr-1;       // ignore FS: to execute regular opcode
+            break;
+
+        case 0xFF:              /* GRP 5 Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            switch((nextop>>3)&7) {
+                case 0:                 /* INC Ed */
+                    ED->dword[0] = inc32(emu, ED->dword[0]);
+                    break;
+                case 1:                 /* DEC Ed */
+                    ED->dword[0] = dec32(emu, ED->dword[0]);
+                    break;
+                case 2:                 /* CALL NEAR Ed */
+                    R_EIP = (uintptr_t)getAlternate((void*)ED->dword[0]);
+                    Push(emu, addr);
+                    addr = R_EIP;
+                    break;
+                case 3:                 /* CALL FAR Ed */
+                    if(nextop>0xc0) {
+                        printf_log(LOG_NONE, "Illegal Opcode %02X %02X\n", opcode, nextop);
+                        emu->quit=1;
+                        emu->error |= ERR_ILLEGAL;
+                        return 0;
+                    } else {
+                        Push16(emu, R_CS);
+                        Push(emu, addr);
+                        addr = ED->dword[0];
+                        R_CS = (ED+1)->word[0];
+                    }
+                    break;
+                case 4:                 /* JMP NEAR Ed */
+                    addr = (uintptr_t)getAlternate((void*)ED->dword[0]);
+                    break;
+                case 5:                 /* JMP FAR Ed */
+                    if(nextop>0xc0) {
+                        printf_log(LOG_NONE, "Illegal Opcode 0x%02X 0x%02X\n", opcode, nextop);
+                        emu->quit=1;
+                        emu->error |= ERR_ILLEGAL;
+                        return 0;
+                    } else {
+                        addr = ED->dword[0];
+                        R_CS = (ED+1)->word[0];
+                    }
+                    break;
+                case 6:                 /* Push Ed */
+                    Push(emu, ED->dword[0]);
+                    break;
+                default:
+                    printf_log(LOG_NONE, "Illegal Opcode 0x%02X 0x%02X\n", opcode, nextop);
+                    emu->quit=1;
+                    emu->error |= ERR_ILLEGAL;
+                    return 0;
+            }
+            break;
+        default:
+            return 0;
+    }
+    return addr;
+}

src/emu/x86run640f.c

@@ -0,0 +1,55 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+#include "bridge.h"
+#ifdef DYNAREC
+#include "../dynarec/arm_lock_helper.h"
+#endif
+
+#include "modrm.h"
+
+
+#ifdef TEST_INTERPRETER
+uintptr_t Test640F(x86test_t *test, uintptr_t tlsdata, uintptr_t addr)
+#else
+uintptr_t Run640F(x86emu_t *emu, uintptr_t tlsdata, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    reg32_t *oped;
+    uint8_t tmp8u;
+    uint32_t tmp32u;
+    int32_t tmp32s;
+    #ifdef TEST_INTERPRETER
+    x86emu_t* emu = test->emu;
+    #endif
+    uint8_t opcode = F8;
+    switch(opcode) {
+
+        case 0xAF:                      /* IMUL Gd,Ed */
+            nextop = F8;
+            GET_ED_OFFS(tlsdata);
+            GD.dword[0] = imul32(emu, GD.dword[0], ED->dword[0]);
+            break;
+
+        case 0xB6:                              /* MOVZX Gd,GS:Eb */
+            nextop = F8;
+            GET_EB_OFFS(tlsdata);
+            GD.dword[0] = EB->byte[0];
+            break;
+
+        default:
+            return 0;
+    }
+    return addr;
+}

src/emu/x86run6466.c

@@ -0,0 +1,55 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+#include "bridge.h"
+#ifdef DYNAREC
+#include "../dynarec/arm_lock_helper.h"
+#endif
+
+#include "modrm.h"
+
+
+#ifdef TEST_INTERPRETER
+uintptr_t Test6466(x86test_t *test, uintptr_t tlsdata, uintptr_t addr)
+#else
+uintptr_t Run6466(x86emu_t *emu, uintptr_t tlsdata, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    reg32_t *oped;
+    uint8_t tmp8u;
+    uint32_t tmp32u;
+    int32_t tmp32s;
+    #ifdef TEST_INTERPRETER
+    x86emu_t* emu = test->emu;
+    #endif
+    uint8_t opcode = F8;
+    switch(opcode) {
+
+        case 0x03:                              /* ADD Gw, FS:Ew */
+            nextop = F8;
+            GET_EW_OFFS(tlsdata);
+            GW.word[0] = add16(emu, GW.word[0], EW->word[0]);
+            break;
+
+        case 0x8B:                              /* MOV Gw,FS:Ew */
+            nextop = F8;
+            GET_EW_OFFS(tlsdata);
+            GW.word[0] = EW->word[0];
+            break;
+
+        default:
+            return 0;
+    }
+    return addr;
+}

src/emu/x86run6467.c

@@ -0,0 +1,92 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+#include "bridge.h"
+#ifdef DYNAREC
+#include "../dynarec/arm_lock_helper.h"
+#endif
+
+#include "modrm.h"
+
+
+#ifdef TEST_INTERPRETER
+uintptr_t Test6467(x86test_t *test, uintptr_t tlsdata, uintptr_t addr)
+#else
+uintptr_t Run6467(x86emu_t *emu, uintptr_t tlsdata, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    reg32_t *oped;
+    uint8_t tmp8u;
+    uint32_t tmp32u;
+    int32_t tmp32s;
+    #ifdef TEST_INTERPRETER
+    x86emu_t* emu = test->emu;
+    #endif
+    uint8_t opcode = F8;
+    switch(opcode) {
+
+        case 0x3B:                              /* CMP GD, FS:Ed16 */
+            nextop = F8;
+            GET_EW16_OFFS(tlsdata);
+            cmp32(emu, GD.dword[0], ED->dword[0]);
+            break;
+
+        case 0x89:                              /* MOV ED16,Gd */
+            nextop = F8;
+            GET_EW16_OFFS(tlsdata);
+            ED->dword[0] = GD.dword[0];
+            break;
+
+        case 0x8B:                              /* MOV Gd,Ed16 */
+            nextop = F8;
+            GET_EW16_OFFS(tlsdata);
+            GD.dword[0] = ED->dword[0];
+            break;
+
+        case 0x8F:                              /* POP FS:Ed */
+            nextop = F8;
+            GET_EW16_OFFS(tlsdata);
+            ED->dword[0] = Pop(emu);
+            break;
+
+        case 0xA1:                              /* MOV EAX,Ov16 */
+            tmp32u = F16;
+            R_EAX = *(uint32_t*)(tlsdata + tmp32u);
+            break;
+        case 0xA3:                              /* MOV Ov16,EAX */
+            tmp32u = F16;
+            #ifdef TEST_INTERPRETER
+            test->memaddr = tlsdata + tmp32u;
+            test->memsize = 4;
+            *(uint32_t*)(test->mem) = R_EAX;
+            #else
+            *(uint32_t*)(tlsdata + tmp32u) = R_EAX;
+            #endif
+            break;
+        case 0xFF:                              /* GRP 5 Ed */
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 6:                         /* Push Ed */
+                    GET_EW16_OFFS(tlsdata);
+                    Push(emu, ED->dword[0]);
+                    break;
+                default:
+                    return 0;
+            }
+            break;
+        default:
+            return 0;
+    }
+    return addr;
+}

src/emu/x86run660f.c

@@ -14,19 +14,14 @@
 #include "x86trace.h"
 #include "box86context.h"
 
-#define F8      *(uint8_t*)(ip++)
-#define F8S     *(int8_t*)(ip++)
-#define F16     *(uint16_t*)(ip+=2, ip-2)
-#define F16S    *(int16_t*)(ip+=2, ip-2)
-#define F32     *(uint32_t*)(ip+=4, ip-4)
-#define F32S    *(int32_t*)(ip+=4, ip-4)
-#define PK(a)   *(uint8_t*)(ip+a)
-
 #include "modrm.h"
 
-void Run660F(x86emu_t *emu)
+#ifdef TEST_INTERPRETER
+uintptr_t Test660F(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t Run660F(x86emu_t *emu, uintptr_t addr)
+#endif
 {
-    uintptr_t ip = R_EIP+2; //skip 66 0F
     uint8_t opcode;
     uint8_t nextop;
     reg32_t *oped;
@@ -43,6 +38,9 @@ void Run660F(x86emu_t *emu)
     #endif
 
 
+    #ifdef TEST_INTERPRETER
+    x86emu_t* emu = test->emu;
+    #endif
     opcode = F8;
     switch(opcode) {
 
@@ -144,8 +142,7 @@ void Run660F(x86emu_t *emu)
                 EW->word[0] = 0;    // dummy operation. Should be CR0 -> Ew
                 break;
             default:
-                ip = R_EIP;
-                UnimpOpcode(emu);
+                return 0;
         }
         break;
     case 0x10:                      /* MOVUPD Gx, Ex */
@@ -160,7 +157,7 @@ void Run660F(x86emu_t *emu)
         break;
     case 0x12:                      /* MOVLPD Gx, Eq */
         nextop = F8;
-        GET_ED;
+        GET_ED8;
         if((uintptr_t)ED & 7)
             memcpy(&GX.q[0], ED, 8);
         else
@@ -168,7 +165,7 @@ void Run660F(x86emu_t *emu)
         break;
     case 0x13:                      /* MOVLPD Eq, Gx */
         nextop = F8;
-        GET_ED;
+        GET_ED8;
         if((uintptr_t)ED & 7)
             memcpy(ED, &GX.q[0], 8);
         else
@@ -187,18 +184,18 @@ void Run660F(x86emu_t *emu)
         break;
     case 0x16:                      /* MOVHPD Gx, Ed */
         nextop = F8;
-        GET_ED;
+        GET_ED8;
         GX.q[1] = *(uint64_t*)ED;
         break;
     case 0x17:                      /* MOVHPD Ed, Gx */
         nextop = F8;
-        GET_ED;
+        GET_ED8;
         *(uint64_t*)ED = GX.q[1];
         break;
 
     case 0x1F:                      /* NOP (multi-byte) */
         nextop = F8;
-        GET_ED;
+        GET_ED_;
         break;
     
     case 0x28:                      /* MOVAPD Gx, Ex */
@@ -219,35 +216,45 @@ void Run660F(x86emu_t *emu)
         GX.d[0] = EM->sd[0];
         GX.d[1] = EM->sd[1];
         break;
-
-
+    case 0x2B:                      /* MOVNTPD Ex, Gx */
+        nextop = F8;
+        GET_EX;
+        EX->q[0] = GX.q[0]; // Address needs to ba aligned for MOVNTPD
+        EX->q[1] = GX.q[1];
+        break;
     case 0x2C:                      /* CVTTPD2PI Gm, Ex */
         nextop = F8;
         GET_EX;
-        GM.sd[0] = EX->d[0];
-        GM.sd[1] = EX->d[1];
+        if(!isfinite(EX->d[0]) || EX->d[0]>(double)0x7fffffff || EX->d[0]<-(double)0x80000000U)
+            GM.ud[0] = 0x80000000;
+        else
+            GM.sd[0] = EX->d[0];
+        if(!isfinite(EX->d[1]) || EX->d[1]>(double)0x7fffffff || EX->d[1]<-(double)0x80000000U)
+            GM.ud[1] = 0x80000000;
+        else
+            GM.sd[1] = EX->d[1];
         break;
     case 0x2D:                      /* CVTPD2PI Gm, Ex */
         nextop = F8;
         GET_EX;
-        switch((emu->mxcsr>>13)&3) {
-            case ROUND_Nearest:
-                GM.sd[0] = floor(EX->d[0]+0.5);
-                GM.sd[1] = floor(EX->d[1]+0.5);
-                break;
-            case ROUND_Down:
-                GM.sd[0] = floor(EX->d[0]);
-                GM.sd[1] = floor(EX->d[1]);
-                break;
-            case ROUND_Up:
-                GM.sd[0] = ceil(EX->d[0]);
-                GM.sd[1] = ceil(EX->d[1]);
-                break;
-            case ROUND_Chop:
-                GM.sd[0] = EX->d[0];
-                GM.sd[1] = EX->d[1];
-                break;
-        }
+        for(int i=0; i<2; ++i)
+            if(!isfinite(EX->d[i]) || EX->d[i]>(double)0x7fffffff || EX->d[i]<-(double)0x80000000U)
+                GM.ud[i] = 0x80000000;
+            else
+                switch(emu->mxcsr.f.MXCSR_RC) {
+                    case ROUND_Nearest:
+                        GM.sd[i] = nearbyint(EX->d[i]);
+                        break;
+                    case ROUND_Down:
+                        GM.sd[i] = floor(EX->d[i]);
+                        break;
+                    case ROUND_Up:
+                        GM.sd[i] = ceil(EX->d[i]);
+                        break;
+                    case ROUND_Chop:
+                        GM.sd[i] = EX->d[i];
+                        break;
+                }
         break;
     case 0x2E:                      /* UCOMISD Gx, Ex */
         // no special check...
@@ -341,7 +348,34 @@ void Run660F(x86emu_t *emu)
                         GX.sw[4+i] = EX->sw[i*2+0] - EX->sw[i*2+1];
                 }
                 break;
-                
+            case 0x06:  /* PHSUBD Gx, Ex */
+                nextop = F8;
+                GET_EX;
+                for (int i=0; i<2; ++i)
+                    GX.sd[i] = GX.sd[i*2+0] - GX.sd[i*2+1];
+                if(&GX == EX) {
+                    GX.q[1] = GX.q[0];
+                } else {
+                    for (int i=0; i<2; ++i)
+                        GX.sd[2+i] = EX->sd[i*2+0] - EX->sd[i*2+1];
+                }
+                break;
+            case 0x07:  /* PHSUBSW Gx, Ex */
+                nextop = F8;
+                GET_EX;
+                for (int i=0; i<4; ++i) {
+                    tmp32s = GX.sw[i*2+0]-GX.sw[i*2+1];
+                    GX.sw[i] = (tmp32s<-32768)?-32768:((tmp32s>32767)?32767:tmp32s);
+                }
+                if(&GX == EX) {
+                    GX.q[1] = GX.q[0];
+                } else {
+                    for (int i=0; i<4; ++i) {
+                        tmp32s = EX->sw[i*2+0] - EX->sw[i*2+1];
+                        GX.sw[4+i] = (tmp32s<-32768)?-32768:((tmp32s>32767)?32767:tmp32s);
+                    }
+                }
+                break;
             case 0x08:  /* PSIGNB Gx, Ex */
                 nextop = F8;
                 GET_EX;
@@ -507,8 +541,7 @@ void Run660F(x86emu_t *emu)
                 break;
 
             default:
-                ip = R_EIP;
-                UnimpOpcode(emu);
+                return 0;
         }
         break;
 
@@ -519,31 +552,31 @@ void Run660F(x86emu_t *emu)
                 nextop = F8;
                 GET_EX;
                 tmp8u = F8;
-                switch((tmp8u & 4) ? ((emu->mxcsr >> 13) & 3) : (tmp8u & 3))
+                switch((tmp8u & 4) ? (emu->mxcsr.f.MXCSR_RC) : (tmp8u & 3))
                 {
                     case ROUND_Nearest:
-                        GX.f[0] = nearbyint(EX->f[0]);
-                        GX.f[1] = nearbyint(EX->f[1]);
-                        GX.f[2] = nearbyint(EX->f[2]);
-                        GX.f[3] = nearbyint(EX->f[3]);
+                        GX.f[0] = nearbyintf(EX->f[0]);
+                        GX.f[1] = nearbyintf(EX->f[1]);
+                        GX.f[2] = nearbyintf(EX->f[2]);
+                        GX.f[3] = nearbyintf(EX->f[3]);
                         break;
                     case ROUND_Down:
-                        GX.f[0] = floor(EX->f[0]);
-                        GX.f[1] = floor(EX->f[1]);
-                        GX.f[2] = floor(EX->f[2]);
-                        GX.f[3] = floor(EX->f[3]);
+                        GX.f[0] = floorf(EX->f[0]);
+                        GX.f[1] = floorf(EX->f[1]);
+                        GX.f[2] = floorf(EX->f[2]);
+                        GX.f[3] = floorf(EX->f[3]);
                         break;
                     case ROUND_Up:
-                        GX.f[0] = ceil(EX->f[0]);
-                        GX.f[1] = ceil(EX->f[1]);
-                        GX.f[2] = ceil(EX->f[2]);
-                        GX.f[3] = ceil(EX->f[3]);
+                        GX.f[0] = ceilf(EX->f[0]);
+                        GX.f[1] = ceilf(EX->f[1]);
+                        GX.f[2] = ceilf(EX->f[2]);
+                        GX.f[3] = ceilf(EX->f[3]);
                         break;
                     case ROUND_Chop:
-                        GX.f[0] = trunc(EX->f[0]);
-                        GX.f[1] = trunc(EX->f[1]);
-                        GX.f[2] = trunc(EX->f[2]);
-                        GX.f[3] = trunc(EX->f[3]);
+                        GX.f[0] = truncf(EX->f[0]);
+                        GX.f[1] = truncf(EX->f[1]);
+                        GX.f[2] = truncf(EX->f[2]);
+                        GX.f[3] = truncf(EX->f[3]);
                         break;
                 }
                 break;
@@ -551,7 +584,7 @@ void Run660F(x86emu_t *emu)
                 nextop = F8;
                 GET_EX;
                 tmp8u = F8;
-                switch((tmp8u & 4) ? ((emu->mxcsr >> 13) & 3) : (tmp8u & 3))
+                switch((tmp8u & 4) ? (emu->mxcsr.f.MXCSR_RC) : (tmp8u & 3))
                 {
                     case ROUND_Nearest:
                         GX.d[0] = nearbyint(EX->d[0]);
@@ -575,19 +608,19 @@ void Run660F(x86emu_t *emu)
                 nextop = F8;
                 GET_EX;
                 tmp8u = F8;
-                switch((tmp8u & 4) ? ((emu->mxcsr >> 13) & 3) : (tmp8u & 3))
+                switch((tmp8u & 4) ? (emu->mxcsr.f.MXCSR_RC) : (tmp8u & 3))
                 {
                     case ROUND_Nearest:
-                        GX.f[0] = nearbyint(EX->f[0]);
+                        GX.f[0] = nearbyintf(EX->f[0]);
                         break;
                     case ROUND_Down:
-                        GX.f[0] = floor(EX->f[0]);
+                        GX.f[0] = floorf(EX->f[0]);
                         break;
                     case ROUND_Up:
-                        GX.f[0] = ceil(EX->f[0]);
+                        GX.f[0] = ceilf(EX->f[0]);
                         break;
                     case ROUND_Chop:
-                        GX.f[0] = trunc(EX->f[0]);
+                        GX.f[0] = truncf(EX->f[0]);
                         break;
                 }
                 break;
@@ -595,7 +628,7 @@ void Run660F(x86emu_t *emu)
                 nextop = F8;
                 GET_EX;
                 tmp8u = F8;
-                switch((tmp8u & 4) ? ((emu->mxcsr >> 13) & 3) : (tmp8u & 3))
+                switch((tmp8u & 4) ? (emu->mxcsr.f.MXCSR_RC) : (tmp8u & 3))
                 {
                     case ROUND_Nearest:
                         GX.d[0] = nearbyint(EX->d[0]);
@@ -628,35 +661,12 @@ void Run660F(x86emu_t *emu)
                 if(tmp8u>31)
                     {GX.q[0] = GX.q[1] = 0;}
                 else
-                #if 0
-                    if(tmp8u>15) {
-                    tmp8u=(tmp8u-16)*8;
-                    if (tmp8u < 64) {
-                        GX.q[0] = (GX.q[0] >> tmp8u) | (GX.q[1] << (64 - tmp8u));
-                        GX.q[1] = (GX.q[1] >> tmp8u);
-                    } else {
-                        GX.q[0] = GX.q[1] >> (tmp8u - 64);
-                        GX.q[1] = 0;
-                    }                    
-                } else {
-                    tmp8u*=8;
-                    if (tmp8u < 64) {
-                        GX.q[0] = (EX->q[0] >> tmp8u) | (EX->q[1] << (64 - tmp8u));
-                        GX.q[1] = (EX->q[1] >> tmp8u) | (GX.q[0] << (64-tmp8u));
-                    } else {
-                        tmp8u -= 64;
-                        GX.q[0] = (EX->q[1] >> tmp8u) | (GX.q[0] << (64 - tmp8u));
-                        GX.q[1] = (GX.q[0] >> tmp8u) | (GX.q[1] >> (64 - tmp8u));
-                    }                    
-                }
-                #else
                 {
                     for (int i=0; i<16; ++i, ++tmp8u)
                         eax1.ub[i] = (tmp8u>15)?((tmp8u>31)?0:GX.ub[tmp8u-16]):EX->ub[tmp8u];
                     GX.q[0] = eax1.q[0];
                     GX.q[1] = eax1.q[1];
                 }
-                #endif
                 break;
 
             case 0x16:      // PEXTRD ED, GX, u8
@@ -690,8 +700,7 @@ void Run660F(x86emu_t *emu)
                 break;
 
             default:
-                ip = R_EIP;
-                UnimpOpcode(emu);
+                return 0;
         }
         break;
 
@@ -775,32 +784,25 @@ void Run660F(x86emu_t *emu)
     case 0x5B:                      /* CVTPS2DQ Gx, Ex */
         nextop = F8;
         GET_EX;
-        switch((emu->mxcsr>>13)&3) {
-            case ROUND_Nearest:
-                GX.sd[0] = floorf(EX->f[0]+0.5f);
-                GX.sd[1] = floorf(EX->f[1]+0.5f);
-                GX.sd[2] = floorf(EX->f[2]+0.5f);
-                GX.sd[3] = floorf(EX->f[3]+0.5f);
-                break;
-            case ROUND_Down:
-                GX.sd[0] = floorf(EX->f[0]);
-                GX.sd[1] = floorf(EX->f[1]);
-                GX.sd[2] = floorf(EX->f[2]);
-                GX.sd[3] = floorf(EX->f[3]);
-                break;
-            case ROUND_Up:
-                GX.sd[0] = ceilf(EX->f[0]);
-                GX.sd[1] = ceilf(EX->f[1]);
-                GX.sd[2] = ceilf(EX->f[2]);
-                GX.sd[3] = ceilf(EX->f[3]);
-                break;
-            case ROUND_Chop:
-                GX.sd[0] = EX->f[0];
-                GX.sd[1] = EX->f[1];
-                GX.sd[2] = EX->f[2];
-                GX.sd[3] = EX->f[3];
-                break;
-        }
+        for(int i=0; i<4; ++i)
+            // note, converting 0x7fffffff to float gives 0x80000000 basicaly, because of the loss or precision in the mantice. So uing >= here...
+            if(isnanf(EX->f[i]) || isinff(EX->f[i]) || EX->f[i]>=(float)0x80000000U || EX->f[i]<-(float)0x80000000U)
+                GX.sd[i] = 0x80000000;
+            else
+                switch(emu->mxcsr.f.MXCSR_RC) {
+                    case ROUND_Nearest:
+                        GX.sd[i] = nearbyintf(EX->f[i]);
+                        break;
+                    case ROUND_Down:
+                        GX.sd[i] = floorf(EX->f[i]);
+                        break;
+                    case ROUND_Up:
+                        GX.sd[i] = ceilf(EX->f[i]);
+                        break;
+                    case ROUND_Chop:
+                        GX.sd[i] = EX->f[i];
+                        break;
+                }
         break;
     case 0x5C:                      /* SUBPD Gx, Ex */
         nextop = F8;
@@ -1018,8 +1020,7 @@ void Run660F(x86emu_t *emu)
                     for (int i=0; i<8; ++i) EX->uw[i] <<= tmp8u;
                 break;
             default:
-                ip = R_EIP;
-                UnimpOpcode(emu);
+                return 0;
         }
         break;
     case 0x72:  /* GRP */
@@ -1045,8 +1046,7 @@ void Run660F(x86emu_t *emu)
                     for (int i=0; i<4; ++i) EX->ud[i] <<= tmp8u;
                 break;
             default:
-                ip = R_EIP;
-                UnimpOpcode(emu);
+                return 0;
         }
         break;
     case 0x73:  /* GRP */
@@ -1098,8 +1098,7 @@ void Run660F(x86emu_t *emu)
                 }
                 break;
             default:
-                ip = R_EIP;
-                UnimpOpcode(emu);
+                return 0;
         }
         break;
     case 0x74:  /* PCMPEQB Gx,Ex */
@@ -1192,7 +1191,12 @@ void Run660F(x86emu_t *emu)
         GET_EW;
         if((nextop&0xC0)!=0xC0)
         {
+            #ifdef TEST_INTERPRETER
+            test->memaddr=((test->memaddr)+tmp32s*2);
+            *(uint16_t*)test->mem = *(uint16_t*)test->memaddr;
+            #else
             EW=(reg32_t*)(((uintptr_t)(EW))+(tmp32s*2));
+            #endif
         }
         if(EW->word[0] & (1<<tmp8u))
             SET_FLAG(F_CF);
@@ -1219,7 +1223,12 @@ void Run660F(x86emu_t *emu)
         GET_EW;
         if((nextop&0xC0)!=0xC0)
         {
+            #ifdef TEST_INTERPRETER
+            test->memaddr=((test->memaddr)+tmp32s*2);
+            *(uint16_t*)test->mem = *(uint16_t*)test->memaddr;
+            #else
             EW=(reg32_t*)(((uintptr_t)(EW))+(tmp32s*2));
+            #endif
         }
         if(EW->word[0] & (1<<tmp8u))
             SET_FLAG(F_CF);
@@ -1265,7 +1274,12 @@ void Run660F(x86emu_t *emu)
         GET_EW;
         if((nextop&0xC0)!=0xC0)
         {
+            #ifdef TEST_INTERPRETER
+            test->memaddr=((test->memaddr)+tmp32s*2);
+            *(uint16_t*)test->mem = *(uint16_t*)test->memaddr;
+            #else
             EW=(reg32_t*)(((uintptr_t)(EW))+(tmp32s*2));
+            #endif
         }
         if(EW->word[0] & (1<<tmp8u)) {
             SET_FLAG(F_CF);
@@ -1334,8 +1348,7 @@ void Run660F(x86emu_t *emu)
                 break;
 
             default:
-                ip = R_EIP;
-                UnimpOpcode(emu);
+                return 0;
         }
         break;
     case 0xBB:                      /* BTC Ew,Gw */
@@ -1347,7 +1360,12 @@ void Run660F(x86emu_t *emu)
         GET_EW;
         if((nextop&0xC0)!=0xC0)
         {
+            #ifdef TEST_INTERPRETER
+            test->memaddr=((test->memaddr)+tmp32s*2);
+            *(uint16_t*)test->mem = *(uint16_t*)test->memaddr;
+            #else
             EW=(reg32_t*)(((uintptr_t)(EW))+(tmp32s*2));
+            #endif
         }
         if(EW->word[0] & (1<<tmp8u))
             SET_FLAG(F_CF);
@@ -1514,8 +1532,7 @@ void Run660F(x86emu_t *emu)
                 if(EX->ub[i]&0x80)
                     GD.dword[0] |= (1<<i);
         } else {
-            ip = R_EIP;
-            UnimpOpcode(emu);
+            return 0;
         }
         break;
     case 0xD8:  /* PSUBUSB Gx,Ex */
@@ -1590,7 +1607,7 @@ void Run660F(x86emu_t *emu)
     case 0xE2:  /* PSRAD Gx, Ex */
         nextop = F8;
         GET_EX;
-        tmp8u=(EX->q[0]>31)?32:EX->ub[0];
+        tmp8u=(EX->q[0]>31)?31:EX->ub[0];
         for (int i=0; i<4; ++i)
             GX.sd[i] >>= tmp8u;
         break;
@@ -1619,8 +1636,14 @@ void Run660F(x86emu_t *emu)
     case 0xE6:  /* CVTTPD2DQ Gx, Ex */
         nextop = F8;
         GET_EX;
-        GX.sd[0] = EX->d[0];
-        GX.sd[1] = EX->d[1];
+        if(isnan(EX->d[0]) || isinf(EX->d[0]) || EX->d[0]>(double)0x7fffffff || EX->d[0]<-(double)0x80000000U)
+            GX.sd[0] = 0x80000000;
+        else
+            GX.sd[0] = EX->d[0];
+        if(isnan(EX->d[1]) || isinf(EX->d[1]) || EX->d[1]>(double)0x7fffffff || EX->d[1]<-(double)0x80000000U)
+            GX.sd[1] = 0x80000000;
+        else
+            GX.sd[1] = EX->d[1];
         GX.q[1] = 0;
         break;
     case 0xE7:   /* MOVNTDQ Ex, Gx */
@@ -1795,10 +1818,8 @@ void Run660F(x86emu_t *emu)
         break;
  
     default:
-        ip = R_EIP;
-        UnimpOpcode(emu);
+        return 0;
     }
 
-    R_EIP = ip;
-    return;
+    return addr;
 }

src/emu/x86run66d9.c

@@ -0,0 +1,103 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t Test66D9(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t Run66D9(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t* emu = test->emu;
+    #endif
+
+    uint8_t opcode = F8;
+    switch(opcode) {
+        
+        case 0xC0:
+        case 0xC1:
+        case 0xC2:
+        case 0xC3:
+        case 0xC4:
+        case 0xC5:
+        case 0xC6:
+        case 0xC7:
+        case 0xC8:
+        case 0xC9:
+        case 0xCA:
+        case 0xCB:
+        case 0xCC:
+        case 0xCD:
+        case 0xCE:
+        case 0xCF:
+        case 0xD0:
+        case 0xE0:
+        case 0xE5:
+        case 0xE8:
+        case 0xE9:
+        case 0xEA:
+        case 0xEB:
+        case 0xEC:
+        case 0xED:
+        case 0xEE:
+        case 0xFC:
+        case 0xE1:
+        case 0xE4:
+        case 0xF0:
+        case 0xF1:
+        case 0xF2:
+        case 0xF3:
+        case 0xF4:
+        case 0xF5:
+        case 0xF6:
+        case 0xF7:
+        case 0xF8:
+        case 0xF9:
+        case 0xFA:
+        case 0xFB:
+        case 0xFD:
+        case 0xFE:
+        case 0xFF:
+            return 0;
+
+        default:
+            switch((nextop>>3)&7) {
+                case 4:     /* FLDENV m */
+                    // warning, incomplete
+                    GET_EW;
+                    #ifndef TEST_INTERPRETER
+                    fpu_loadenv(emu, (char*)ED, 1);
+                    #endif
+                    break;
+                case 6:     /* FNSTENV m */
+                    // warning, incomplete
+                    GET_EW;
+                    #ifndef TEST_INTERPRETER
+                    fpu_savenv(emu, (char*)ED, 1);
+                    #endif
+                    break;
+                default:
+                    return 0;
+            }
+    }
+
+    return addr;
+}

src/emu/x86run66dd.c

@@ -0,0 +1,122 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t Test66DD(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t Run66DD(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t* emu = test->emu;
+    #endif
+
+    uint8_t opcode = F8;
+    switch(opcode) {
+
+        case 0xC0:
+        case 0xC1:
+        case 0xC2:
+        case 0xC3:
+        case 0xC4:
+        case 0xC5:
+        case 0xC6:
+        case 0xC7:
+        case 0xC8:
+        case 0xC9:
+        case 0xCA:
+        case 0xCB:
+        case 0xCC:
+        case 0xCD:
+        case 0xCE:
+        case 0xCF:
+        case 0xD0:
+        case 0xE0:
+        case 0xE5:
+        case 0xE8:
+        case 0xE9:
+        case 0xEA:
+        case 0xEB:
+        case 0xEC:
+        case 0xED:
+        case 0xEE:
+        case 0xFC:
+        case 0xE1:
+        case 0xE4:
+        case 0xF0:
+        case 0xF1:
+        case 0xF2:
+        case 0xF3:
+        case 0xF4:
+        case 0xF5:
+        case 0xF6:
+        case 0xF7:
+        case 0xF8:
+        case 0xF9:
+        case 0xFA:
+        case 0xFB:
+        case 0xFD:
+        case 0xFE:
+        case 0xFF:
+            return 0;
+        break;
+
+        default:
+            switch((nextop>>3)&7) {
+                case 4: /* FRSTOR m94byte */
+                    GET_EW;
+                    #ifndef TEST_INTERPRETER
+                    fpu_loadenv(emu, (char*)ED, 1);
+                    // get the STx
+                    {
+                        char* p =(char*)ED;
+                        p += 14;
+                        for (int i=0; i<8; ++i) {
+                            LD2D(p, &ST(i).d);
+                            p+=10;
+                        }
+                    }
+                    #endif
+                    break;
+                case 6: /* FNSAVE m94byte */
+                    GET_EW;
+                    // ENV first...
+                    #ifndef TEST_INTERPRETER
+                    fpu_savenv(emu, (char*)ED, 1);
+                    // save the STx
+                    {
+                        char* p =(char*)ED;
+                        p += 14;
+                        for (int i=0; i<8; ++i) {
+                            D2LD(&ST(i).d, p);
+                            p+=10;
+                        }
+                    }
+                    #endif
+                    reset_fpu(emu);
+                    break;
+                default:
+                    return 0;
+            }
+    }
+
+    return addr;
+}

src/emu/x86run67.c

@@ -0,0 +1,147 @@
+#define _GNU_SOURCE
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "x87emu_private.h"
+#include "box86context.h"
+#include "my_cpuid.h"
+#include "bridge.h"
+#include "signals.h"
+#ifdef DYNAREC
+#include "../dynarec/arm_lock_helper.h"
+#endif
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t Test67(x86test_t *test, int rep, uintptr_t addr)
+#else
+uintptr_t Run67(x86emu_t *emu, int rep, uintptr_t addr)
+#endif
+{
+    uint8_t opcode;
+    uint8_t nextop;
+    reg32_t *oped;
+    uint8_t tmp8u, tmp8u2;
+    int8_t tmp8s;
+    uint16_t tmp16u, tmp16u2;
+    int16_t tmp16s;
+    uint32_t tmp32u, tmp32u2, tmp32u3;
+    int32_t tmp32s, tmp32s2;
+    uint64_t tmp64u;
+    int64_t tmp64s;
+    double d;
+    float f;
+    int64_t ll;
+    sse_regs_t *opex, eax1;
+    mmx87_regs_t *opem, eam1;
+    uintptr_t tlsdata;
+
+#ifdef TEST_INTERPRETER
+    x86emu_t *emu = test->emu;
+#endif
+    opcode = F8;
+
+    while(opcode==0x67)
+        opcode = F8;
+
+    while((opcode==0xF2) || (opcode==0xF3)) {
+        rep = opcode-0xF1;
+        opcode = F8;
+    }
+
+    switch(opcode) {
+
+        case 0x64:
+            tlsdata = GetFSBaseEmu(emu);
+            #ifdef TEST_INTERPRETER
+            return Test6467(test, tlsdata, addr);
+            #else
+            return Run6467(emu, tlsdata, addr);
+            #endif
+
+        case 0x66:                      /* MoooRE opcodes */
+            #ifdef TEST_INTERPRETER
+            return Test6766(test, rep, addr);
+            #else
+            return Run6766(emu, rep, addr);
+            #endif
+
+        case 0x6C:                      /* INSB */
+            tmp32u = rep?R_CX:1;
+            while(tmp32u--) {
+                #ifdef TEST_INTERPRETER
+                *(int8_t*)(R_DI+GetESBaseEmu(emu)) = 0;         // faking port read, using actual segment ES, just in case
+                #endif
+                if(ACCESS_FLAG(F_DF))
+                    R_DI-=1;
+                else
+                    R_DI+=1;
+            }
+            if(rep) R_CX = 0;
+            break;
+
+        case 0xAC:                      /* LODSB */
+            tmp32u = rep?R_CX:1;
+            while(tmp32u--) {
+                R_AL = *(int8_t*)(R_SI+GetDSBaseEmu(emu));
+                if(ACCESS_FLAG(F_DF))
+                    R_SI-=1;
+                else
+                    R_SI+=1;
+            }
+            if(rep) R_CX = 0;
+            break;
+
+        case 0xE0:                      /* LOOPNZ */
+            CHECK_FLAGS(emu);
+            tmp8s = F8S;
+            --R_CX; // don't update flags
+            if(R_CX && !ACCESS_FLAG(F_ZF))
+                addr += tmp8s;
+            break;
+        case 0xE1:                      /* LOOPZ */
+            CHECK_FLAGS(emu);
+            tmp8s = F8S;
+            --R_CX; // don't update flags
+            if(R_CX && ACCESS_FLAG(F_ZF))
+                addr += tmp8s;
+            break;
+        case 0xE2:                      /* LOOP */
+            tmp8s = F8S;
+            --R_CX; // don't update flags
+            if(R_CX)
+                addr += tmp8s;
+            break;
+        case 0xE3:                      /* JCXZ */
+            tmp8s = F8S;
+            if(!R_CX)
+                addr += tmp8s;
+            break;
+
+        case 0xE8:                      /* CALL Id */
+            tmp32s = F32S; // call is relative
+            Push(emu, addr);
+            addr += tmp32s;
+            break;
+
+
+        default:
+            return 0;
+    }
+    return addr;
+}

src/emu/x86run6766.c

@@ -0,0 +1,58 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+#include "bridge.h"
+#ifdef DYNAREC
+#include "../dynarec/arm_lock_helper.h"
+#endif
+
+#include "modrm.h"
+
+
+#ifdef TEST_INTERPRETER
+uintptr_t Test6766(x86test_t *test, int rep, uintptr_t addr)
+#else
+uintptr_t Run6766(x86emu_t *emu, int rep, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    reg32_t *oped;
+    uint8_t tmp8u;
+    uint32_t tmp32u;
+    int32_t tmp32s;
+    #ifdef TEST_INTERPRETER
+    x86emu_t* emu = test->emu;
+    #endif
+    uint8_t opcode = F8;
+
+    while((opcode==0x2E) || (opcode==0x66))   // ignoring CS: or multiple 0x66
+        opcode = F8;
+
+    while((opcode==0xF2) || (opcode==0xF3)) {
+        rep = opcode-0xF1;
+        opcode = F8;
+    }
+
+    switch(opcode) {
+
+        case 0x8D:                              /* LEA Gw,Ew */
+            nextop = F8;
+            GET_EW16_;
+            GW.word[0] = (uint16_t)(uintptr_t)ED;
+            break;
+
+        default:
+            return 0;
+    }
+    return addr;
+}

src/emu/x86run_private.c

@@ -50,6 +50,7 @@ void EXPORT my___libc_init(x86emu_t* emu, void* raw_args __unused, void (*onexit
 #else
 int32_t EXPORT my___libc_start_main(x86emu_t* emu, int *(main) (int, char * *, char * *), int argc, char * * ubp_av, void (*init) (void), void (*fini) (void), void (*rtld_fini) (void), void (* stack_end))
 {
+    (void)argc; (void)ubp_av; (void)fini; (void)rtld_fini; (void)stack_end;
     // let's cheat and set all args...
     Push(emu, (uint32_t)my_context->envv);
     Push(emu, (uint32_t)my_context->argv);
@@ -818,7 +819,7 @@ uintptr_t evalED(x86emu_t* emu, uintptr_t ip) {
             if((nextop&7)==4) {
                 uint8_t sib = PK(0);
                 ++ip;
-                uintptr_t base = ((sib&0x7)==5)?(PK32(0)):(emu->regs[(sib&0x7)].dword[0]);
+                uintptr_t base = ((sib&0x7)==5)?(uint32_t)(PK32(0)):(emu->regs[(sib&0x7)].dword[0]);
                 base += (emu->sbiidx[(sib>>3)&7]->sdword[0] << (sib>>6)); \
                 return *(uintptr_t*)base;
             } else if((nextop&7)==5) {
@@ -852,7 +853,7 @@ void PrintTrace(x86emu_t* emu, uintptr_t ip, int dynarec)
             (trace_end == 0) 
             || ((ip >= trace_start) && (ip < trace_end))) ) {
         int tid = syscall(SYS_gettid);
-        pthread_mutex_lock(&my_context->mutex_trace);
+        mutex_lock(&my_context->mutex_trace);
 #ifdef DYNAREC
         if((my_context->trace_tid != tid) || (my_context->trace_dynarec!=dynarec)) {
             printf_log(LOG_NONE, "Thread %04d| (%s) |\n", tid, dynarec?"dyn":"int");
@@ -895,7 +896,7 @@ void PrintTrace(x86emu_t* emu, uintptr_t ip, int dynarec)
             }
             printf_log(LOG_NONE, "\n");
         }
-        pthread_mutex_unlock(&my_context->mutex_trace);
+        mutex_unlock(&my_context->mutex_trace);
     }
 }
 

src/emu/x86run_private.h

@@ -4,6 +4,7 @@
 #include <stdint.h>
 #include "regs.h"
 #include "x86emu_private.h"
+#include "box86context.h"
 typedef struct x86emu_s x86emu_t;
 
 static inline uint8_t Fetch8(x86emu_t *emu) {return *(uint8_t*)(R_EIP++);}
@@ -41,15 +42,25 @@ static inline uint32_t Pop(x86emu_t *emu)
     return *st;
 }
 
+#ifdef TEST_INTERPRETER
+#define Push(E, V)  do{E->regs[_SP].dword[0] -=4; test->memsize = 4; *(uint64_t*)test->mem = (V); test->memaddr = E->regs[_SP].dword[0];}while(0)
+#define Push16(E, V)  do{E->regs[_SP].dword[0] -=2; test->memsize = 2; *(uint16_t*)test->mem = (V); test->memaddr = E->regs[_SP].dword[0];}while(0)
+#else
 static inline void Push(x86emu_t *emu, uint32_t v)
 {
     R_ESP -= 4;
     *((uint32_t*)R_ESP) = v;
 }
+#endif
 
+static inline void PushExit(x86emu_t* emu)
+{
+    R_ESP -= 4;
+    *((uint32_t*)R_ESP) = my_context->exit_bridge;
+}
 
+#if 0
 // the op code definition can be found here: http://ref.x86asm.net/geek32.html
-
 static inline reg32_t* GetECommon(x86emu_t* emu, uint32_t m)
 {
     if (m<=7) {
@@ -188,23 +199,69 @@ static inline sse_regs_t* GetGx(x86emu_t *emu, uint32_t v)
     uint8_t m = (v&0x38)>>3;
     return &emu->xmm[m&7];
 }
-
+#endif
 void UpdateFlags(x86emu_t *emu);
 
 #define CHECK_FLAGS(emu) if(emu->df) UpdateFlags(emu)
 #define RESET_FLAGS(emu) emu->df = d_none
 
-void Run67(x86emu_t *emu);
-void Run0F(x86emu_t *emu);
-void Run660F(x86emu_t *emu);
-void Run66D9(x86emu_t *emu);    // x87
-void Run6766(x86emu_t *emu);
-void RunGS(x86emu_t *emu);
-void RunGS0F(x86emu_t *emu, uintptr_t tlsdata);
-void RunFS(x86emu_t *emu);
-void RunFS66(x86emu_t *emu, uintptr_t tlsdata);
-void RunLock(x86emu_t *emu);
-void RunLock66(x86emu_t *emu);
+uintptr_t Run0F(x86emu_t *emu, uintptr_t addr, int *step);
+uintptr_t Run64(x86emu_t *emu, int seg, uintptr_t addr);
+uintptr_t Run640F(x86emu_t *emu, uintptr_t tlsdata, uintptr_t addr);
+uintptr_t Run6466(x86emu_t *emu, uintptr_t tlsdata, uintptr_t addr);
+uintptr_t Run6467(x86emu_t *emu, uintptr_t tlsdata, uintptr_t addr);
+uintptr_t Run66(x86emu_t *emu, int rep, uintptr_t addr);
+uintptr_t Run660F(x86emu_t *emu, uintptr_t addr);
+uintptr_t Run6664(x86emu_t *emu, int seg, uintptr_t addr);
+uintptr_t Run66D9(x86emu_t *emu, uintptr_t addr);
+uintptr_t Run66DD(x86emu_t *emu, uintptr_t addr);
+uintptr_t Run66F0(x86emu_t *emu, uintptr_t addr);
+uintptr_t Run67(x86emu_t *emu, int rep, uintptr_t addr);
+uintptr_t Run670F(x86emu_t *emu, int rep, uintptr_t addr);
+uintptr_t Run6766(x86emu_t *emu, int rep, uintptr_t addr);
+uintptr_t Run67660F(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunD8(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunD9(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunDA(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunDB(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunDC(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunDD(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunDE(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunDF(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunF0(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunF066(x86emu_t *emu, uintptr_t addr);
+uintptr_t RunF20F(x86emu_t *emu, uintptr_t addr, int *step);
+uintptr_t RunF30F(x86emu_t *emu, uintptr_t addr);
+
+#ifdef DYNAREC
+uintptr_t Test0F(x86test_t *test, uintptr_t addr, int *step);
+uintptr_t Test64(x86test_t *test, int seg, uintptr_t addr);
+uintptr_t Test640F(x86test_t *test, uintptr_t tlsdata, uintptr_t addr);
+uintptr_t Test6466(x86test_t *test, uintptr_t tlsdata, uintptr_t addr);
+uintptr_t Test6467(x86test_t *test, uintptr_t tlsdata, uintptr_t addr);
+uintptr_t Test66(x86test_t *test, int rep, uintptr_t addr);
+uintptr_t Test660F(x86test_t *test, uintptr_t addr);
+uintptr_t Test6664(x86test_t *test, int seg, uintptr_t addr);
+uintptr_t Test66D9(x86test_t *test, uintptr_t addr);
+uintptr_t Test66DD(x86test_t *test, uintptr_t addr);
+uintptr_t Test66F0(x86test_t *test, uintptr_t addr);
+uintptr_t Test67(x86test_t *test, int rep, uintptr_t addr);
+uintptr_t Test670F(x86test_t *test, int rep, uintptr_t addr);
+uintptr_t Test6766(x86test_t *test, int rep, uintptr_t addr);
+uintptr_t Test67660F(x86test_t *test, uintptr_t addr);
+uintptr_t TestD8(x86test_t *test, uintptr_t addr);
+uintptr_t TestD9(x86test_t *test, uintptr_t addr);
+uintptr_t TestDA(x86test_t *test, uintptr_t addr);
+uintptr_t TestDB(x86test_t *test, uintptr_t addr);
+uintptr_t TestDC(x86test_t *test, uintptr_t addr);
+uintptr_t TestDD(x86test_t *test, uintptr_t addr);
+uintptr_t TestDE(x86test_t *test, uintptr_t addr);
+uintptr_t TestDF(x86test_t *test, uintptr_t addr);
+uintptr_t TestF0(x86test_t *test, uintptr_t addr);
+uintptr_t TestF066(x86test_t *test, uintptr_t addr);
+uintptr_t TestF20F(x86test_t *test, uintptr_t addr, int *step);
+uintptr_t TestF30F(x86test_t *test, uintptr_t addr);
+#endif
 
 void x86Syscall(x86emu_t *emu);
 void x86Int3(x86emu_t* emu);

src/emu/x86rund8.c

@@ -1,7 +1,40 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wstrict-aliasing"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestD8(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunD8(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    float f;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     nextop = F8;
-    switch(nextop) {
+    switch (nextop) {
         case 0xC0:
         case 0xC1:
         case 0xC2:
@@ -159,7 +192,9 @@
                 }
                 break;
             default:
-                goto _default;
+                return 0;
         }
     }
+   return addr;
+}
 #pragma GCC diagnostic pop

src/emu/x86rund9.c

@@ -1,3 +1,38 @@
+#define _GNU_SOURCE
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestD9(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunD9(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    int32_t tmp32s;
+    int64_t ll;
+    float f;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     nextop = F8;
     switch (nextop) {
         case 0xC0:
@@ -206,7 +241,7 @@
         case 0xE6:
         case 0xE7:
         case 0xEF:
-            goto _default;
+            return 0;
         default:
         switch((nextop>>3)&7) {
             case 0:     /* FLD ST0, Ed float */
@@ -241,7 +276,9 @@
             case 4:     /* FLDENV m */
                 // warning, incomplete
                 GET_ED;
+                #ifndef TEST_INTERPRETER
                 fpu_loadenv(emu, (char*)ED, 0);
+                #endif
                 break;
             case 5:     /* FLDCW Ew */
                 GET_EW;
@@ -251,7 +288,9 @@
             case 6:     /* FNSTENV m */
                 // warning, incomplete
                 GET_ED;
+                #ifndef TEST_INTERPRETER
                 fpu_savenv(emu, (char*)ED, 0);
+                #endif
                 // intruction pointer: 48bits
                 // data (operand) pointer: 48bits
                 // last opcode: 11bits save: 16bits restaured (1st and 2nd opcode only)
@@ -261,6 +300,8 @@
                 EW->word[0] = emu->cw.x16;
                 break;
             default:
-                goto _default;
+                return 0;
         }
     }
+   return addr;
+}

src/emu/x86runda.c

@@ -1,3 +1,37 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestDA(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunDA(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    int32_t tmp32s;
+    int64_t ll;
+    float f;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     nextop = F8;
     switch(nextop) {
 
@@ -66,8 +100,7 @@
     case 0xF8:
     case 0xF9:
     case 0xFD:
-        UnimpOpcode(emu);
-        break;
+        return 0;
     default:
         switch((nextop>>3)&7) {
             case 0:     /* FIADD ST0, Ed int */
@@ -105,3 +138,5 @@
                 break;
         }
     }
+    return addr;
+}

src/emu/x86rundb.c

@@ -1,3 +1,37 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestDB(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunDB(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    int32_t tmp32s;
+    int64_t ll;
+    float f;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     nextop = F8;
     switch(nextop) {
     case 0xC0:      /* FCMOVNB ST(0), ST(i) */
@@ -93,7 +127,7 @@
     case 0xE5:
     case 0xE6:
     case 0xE7:
-        goto _default;
+        return 0;
     default:
         switch((nextop>>3)&7) {
             case 0: /* FILD ST0, Ed */
@@ -103,15 +137,16 @@
                 break;
             case 1: /* FISTTP Ed, ST0 */
                 GET_ED;
-                tmp32s = ST0.d; // TODO: Handling of FPU Exception
-                if(tmp32s==0x7fffffff && isgreater(ST0.d, (double)(int32_t)0x7fffffff))
+                if(isgreater(ST0.d, (double)0x7fffffff) || isless(ST0.d, -(double)0x80000000U) || !isfinite(ST0.d))
                     tmp32s = 0x80000000;
+                else
+                    tmp32s = ST0.d; // TODO: Handling of FPU Exception
                 fpu_do_pop(emu);
                 ED->sdword[0] = tmp32s;
                 break;
             case 2: /* FIST Ed, ST0 */
                 GET_ED;
-                if(isgreater(ST0.d, (double)(int32_t)0x7fffffff) || isless(ST0.d, -(double)(int32_t)0x7fffffff) || !isfinite(ST0.d))
+                if(isgreater(ST0.d, (double)0x7fffffff) || isless(ST0.d, -(double)0x80000000U) || !isfinite(ST0.d))
                     ED->sdword[0] = 0x80000000;
                 else {
                     volatile int32_t tmp = fpu_round(emu, ST0.d);    // tmp to avoid BUS ERROR
@@ -120,7 +155,7 @@
                 break;
             case 3: /* FISTP Ed, ST0 */
                 GET_ED;
-                if(isgreater(ST0.d, (double)(int32_t)0x7fffffff) || isless(ST0.d, -(double)(int32_t)0x7fffffff) || !isfinite(ST0.d))
+                if(isgreater(ST0.d, (double)0x7fffffff) || isless(ST0.d, -(double)0x80000000U) || !isfinite(ST0.d))
                     ED->sdword[0] = 0x80000000;
                 else {
                     volatile int32_t tmp = fpu_round(emu, ST0.d);    // tmp to avoid BUS ERROR
@@ -129,14 +164,14 @@
                 fpu_do_pop(emu);
                 break;
             case 5: /* FLD ST0, Et */
-                GET_ED;
+                GET_EDT;
                 fpu_do_push(emu);
                 memcpy(&STld(0).ld, ED, 10);
                 LD2D(&STld(0).ld, &ST(0).d);
                 STld(0).uref = ST0.q;
                 break;
             case 7: /* FSTP tbyte */
-                GET_ED;
+                GET_EDT;
                 if(ST0.q!=STld(0).uref)
                     D2LD(&ST0.d, ED);
                 else
@@ -144,6 +179,8 @@
                 fpu_do_pop(emu);
                 break;
             default:
-                goto _default;
+                return 0;
         }
     }
+    return addr;
+}

src/emu/x86rundc.c

@@ -1,3 +1,37 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestDC(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunDC(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    int32_t tmp32s;
+    int64_t ll;
+    double d;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     nextop = F8;
     switch(nextop) {
         case 0xC0:
@@ -82,7 +116,7 @@
             ST(nextop&7).d /=  ST0.d;
             break;
         default:
-            GET_ED;
+            GET_ED8;
             switch((nextop>>3)&7) {
             case 0:         /* FADD ST0, double */
                 if(!(((uintptr_t)ED)&7))
@@ -150,7 +184,8 @@
                 }
                 break;
             default:
-                goto _default;
+                return 0;
         }
     }
-#pragma GCC diagnostic pop
+    return addr;
+}

src/emu/x86rundd.c

@@ -1,3 +1,37 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestDD(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunDD(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    int32_t tmp32s;
+    int64_t ll;
+    double d;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     nextop = F8;
     switch(nextop) {
     
@@ -79,12 +113,12 @@
     case 0xFD:
     case 0xFE:
     case 0xFF:
-        goto _default;
+        return 0;
 
     default:
         switch((nextop>>3)&7) {
             case 0: /* FLD double */
-                GET_ED;
+                GET_ED8;
                 fpu_do_push(emu);
                 if(!(((uintptr_t)ED)&7))
                     ST0.d = *(double*)ED;
@@ -93,17 +127,32 @@
                 }
                 break;
             case 1: /* FISTTP ED qword */
-                GET_ED;
-                if(!(((uintptr_t)ED)&7))
-                    *(int64_t*)ED = ST0.d;
-                else {
-                    int64_t i64 = ST0.d;
+                GET_ED8;
+                if(!(((uintptr_t)ED)&7)) {
+                    if(STll(0).sref==ST(0).sq)
+                        *(int64_t*)ED = STll(0).sq;
+                    else {
+                        if(isgreater(ST0.d, (double)0x7fffffffffffffffLL) || isless(ST0.d, -(double)0x8000000000000000LL) || !isfinite(ST0.d))
+                            *(uint64_t*)ED = 0x8000000000000000LL;
+                        else
+                            *(int64_t*)ED = ST0.d;
+                    }
+                } else {
+                    int64_t i64;
+                    if(STll(0).sref==ST(0).sq)
+                        i64 = STll(0).sq;
+                    else {
+                        if(isgreater(ST0.d, (double)0x7fffffffffffffffLL) || isless(ST0.d, -(double)0x8000000000000000LL) || !isfinite(ST0.d))
+                            i64 = 0x8000000000000000LL;
+                        else
+                            i64 = ST0.d;
+                    }
                     memcpy(ED, &i64, sizeof(int64_t));
                 }
                 fpu_do_pop(emu);
                 break;
             case 2: /* FST double */
-                GET_ED;
+                GET_ED8;
                 if(!(((uintptr_t)ED)&7))
                     *(double*)ED = ST0.d;
                 else {
@@ -111,7 +160,7 @@
                 }
                 break;
             case 3: /* FSTP double */
-                GET_ED;
+                GET_ED8;
                 if(!(((uintptr_t)ED)&7))
                     *(double*)ED = ST0.d;
                 else {
@@ -120,8 +169,10 @@
                 fpu_do_pop(emu);
                 break;
             case 4: /* FRSTOR m108byte */
-                GET_ED;
+                GET_ED_;
+                #ifndef TEST_INTERPRETER
                 fpu_loadenv(emu, (char*)ED, 0);
+                #endif
                 // get the STx
                 {
                     char* p =(char*)ED;
@@ -133,9 +184,10 @@
                 }
                 break;
             case 6: /* FNSAVE m108byte */
-                GET_ED;
+                GET_ED_;
                 // ENV first...
                 // warning, incomplete
+                #ifndef TEST_INTERPRETER
                 fpu_savenv(emu, (char*)ED, 0);
                 // save the STx
                 {
@@ -146,6 +198,7 @@
                         p+=10;
                     }
                 }
+                #endif
                 reset_fpu(emu);
                 break;
             case 7: /* FNSTSW m2byte */
@@ -154,6 +207,8 @@
                 *(uint16_t*)ED = emu->sw.x16;
                 break;
             default:
-                goto _default;
+                return 0;
         }
     }
+    return addr;
+}

src/emu/x86runde.c

@@ -1,3 +1,37 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestDE(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunDE(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    int32_t tmp32s;
+    int64_t ll;
+    double d;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     nextop = F8;
     switch (nextop) {
     case 0xC0:  /* FADDP STx, ST0 */
@@ -95,7 +129,7 @@
     case 0xDD:
     case 0xDE:
     case 0xDF:
-        goto _default;
+        return 0;
     
     default:
         switch((nextop>>3)&7) {
@@ -124,6 +158,8 @@
                 ST0.d = (double)EW->sword[0] / ST0.d;
                 break;
         default:
-            goto _default;
+            return 0;
         }
     }
+    return addr;
+}

src/emu/x86rundf.c

@@ -1,3 +1,37 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x87emu_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "box86context.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestDF(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunDF(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t nextop;
+    int16_t tmp16s;
+    int64_t tmp64s;
+    double d;
+    reg32_t *oped;
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     nextop = F8;
     switch (nextop) {
     case 0xC0:  /* FFREEP STx */
@@ -25,8 +59,11 @@
         break;
 
     case 0xE0:  /* FNSTSW AX */
-        emu->sw.f.F87_TOP = emu->top&7;
-        R_AX = emu->sw.x16;
+        {
+            x87flags_t sw = emu->sw;
+            sw.f.F87_TOP = emu->top&7;
+            R_AX = sw.x16;
+        }
         break;
 
     case 0xE8:  /* FUCOMIP ST0, STx */
@@ -84,7 +121,7 @@
     case 0xFD:
     case 0xFE:
     case 0xFF:
-        goto _default;
+        return 0;
 
     default:
         switch((nextop>>3)&7) {
@@ -96,32 +133,35 @@
             break;
         case 1: /* FISTTP Ew, ST0 */
             GET_EW;
-            tmp16s = ST0.d;
+            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x8000) || !isfinite(ST0.d))
+                tmp16s = 0x8000;
+            else
+                tmp16s = ST0.d;
             EW->sword[0] = tmp16s;
             fpu_do_pop(emu);
             break;
         case 2: /* FIST Ew, ST0 */
             GET_EW;
-            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x7fff) || !isfinite(ST0.d))
+            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x8000) || !isfinite(ST0.d))
                 EW->sword[0] = 0x8000;
             else
                 EW->sword[0] = fpu_round(emu, ST0.d);
             break;
         case 3: /* FISTP Ew, ST0 */
             GET_EW;
-            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x7fff) || !isfinite(ST0.d))
+            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x8000) || !isfinite(ST0.d))
                 EW->sword[0] = 0x8000;
             else
                 EW->sword[0] = fpu_round(emu, ST0.d);
             fpu_do_pop(emu);
             break;
         case 4: /* FBLD ST0, tbytes */
-            GET_ED;
+            GET_EDT;
             fpu_do_push(emu);
             fpu_fbld(emu, (uint8_t*)ED);
             break;
         case 5: /* FILD ST0, Gq */
-            GET_ED;
+            GET_ED8;
             tmp64s = *(int64_t*)ED;
             fpu_do_push(emu);
             ST0.d = tmp64s;
@@ -129,19 +169,19 @@
             STll(0).sref = ST0.sq;
             break;
         case 6: /* FBSTP tbytes, ST0 */
-            GET_ED;
+            GET_EDT;
             fpu_fbst(emu, (uint8_t*)ED);
             fpu_do_pop(emu);
             break;
         case 7: /* FISTP i64 */
-            GET_ED;
+            GET_ED8;
             if((uintptr_t)ED & 0x7) {
                 // un-aligned!
                 if(STll(0).sref==ST(0).sq)
                     memcpy(ED, &STll(0).sq, sizeof(int64_t));
                 else {
                     int64_t i64;
-                    if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, -(double)(int64_t)0x7fffffffffffffffLL) || !isfinite(ST0.d))
+                    if(isgreater(ST0.d, (double)0x7fffffffffffffffLL) || isless(ST0.d, -(double)0x8000000000000000LL) || !isfinite(ST0.d))
                         i64 = 0x8000000000000000LL;
                     else
                         i64 = fpu_round(emu, ST0.d);
@@ -151,7 +191,7 @@
                 if(STll(0).sref==ST(0).sq)
                     *(int64_t*)ED = STll(0).sq;
                 else {
-                    if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, -(double)(int64_t)0x7fffffffffffffffLL) || !isfinite(ST0.d))
+                    if(isgreater(ST0.d, (double)0x7fffffffffffffffLL) || isless(ST0.d, -(double)0x8000000000000000LL) || !isfinite(ST0.d))
                         *(int64_t*)ED = 0x8000000000000000LL;
                     else
                         *(int64_t*)ED = fpu_round(emu, ST0.d);
@@ -160,6 +200,8 @@
             fpu_do_pop(emu);
             break;
         default:
-            goto _default;
+            return 0;
         }
     }
+    return addr;
+}

src/emu/x86runf0.c

@@ -0,0 +1,625 @@
+#define _GNU_SOURCE
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "x87emu_private.h"
+#include "box86context.h"
+#include "my_cpuid.h"
+#include "bridge.h"
+#include "signals.h"
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+#include "../dynarec/arm_lock_helper.h"
+#endif
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestF0(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunF0(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t opcode;
+    uint8_t nextop;
+    reg32_t *oped;
+    uint8_t tmp8u, tmp8u2;
+    int8_t tmp8s;
+    uint16_t tmp16u, tmp16u2;
+    int16_t tmp16s;
+    uint32_t tmp32u, tmp32u2, tmp32u3;
+    int32_t tmp32s, tmp32s2;
+    uint64_t tmp64u;
+    int64_t tmp64s;
+    double d;
+    float f;
+    int64_t ll;
+    sse_regs_t *opex, eax1;
+    mmx87_regs_t *opem, eam1;
+
+#ifdef TEST_INTERPRETER
+    x86emu_t *emu = test->emu;
+#endif
+    opcode = F8;
+
+    switch(opcode) {
+
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+        #define GO(B, OP)                      \
+        case B+0: \
+            nextop = F8;               \
+            GET_EB;             \
+            do {                \
+            tmp8u = arm_lock_read_b(EB);     \
+            tmp8u = OP##8(emu, tmp8u, GB);  \
+            } while (arm_lock_write_b(EB, tmp8u));   \
+            break;                              \
+        case B+1: \
+            nextop = F8;               \
+            GET_ED;             \
+            do {                \
+            tmp32u = arm_lock_read_d(ED);     \
+            tmp32u = OP##32(emu, tmp32u, GD.dword[0]);  \
+            } while (arm_lock_write_d(ED, tmp32u));   \
+            break;                              \
+        case B+2: \
+            nextop = F8;               \
+            GET_EB;                   \
+            GB = OP##8(emu, GB, EB->byte[0]); \
+            break;                              \
+        case B+3: \
+            nextop = F8;               \
+            GET_ED;         \
+            GD.dword[0] = OP##32(emu, GD.dword[0], ED->dword[0]); \
+            break;                              \
+        case B+4: \
+            R_AL = OP##8(emu, R_AL, F8); \
+            break;                              \
+        case B+5: \
+            R_EAX = OP##32(emu, R_EAX, F32); \
+            break;
+#else
+        #define GO(B, OP)                      \
+        case B+0: \
+            nextop = F8;               \
+            GET_EB;             \
+            pthread_mutex_lock(&emu->context->mutex_lock);\
+            EB->byte[0] = OP##8(emu, EB->byte[0], GB);  \
+            pthread_mutex_unlock(&emu->context->mutex_lock);\
+            break;                              \
+        case B+1: \
+            nextop = F8;               \
+            GET_ED;             \
+            pthread_mutex_lock(&emu->context->mutex_lock);\
+            ED->dword[0] = OP##32(emu, ED->dword[0], GD.dword[0]); \
+            pthread_mutex_unlock(&emu->context->mutex_lock);\
+            break;                              \
+        case B+2: \
+            nextop = F8;               \
+            GET_EB;                   \
+            pthread_mutex_lock(&emu->context->mutex_lock);\
+            GB = OP##8(emu, GB, EB->byte[0]); \
+            pthread_mutex_unlock(&emu->context->mutex_lock);\
+            break;                              \
+        case B+3: \
+            nextop = F8;               \
+            GET_ED;         \
+            pthread_mutex_lock(&emu->context->mutex_lock);\
+            GD.dword[0] = OP##32(emu, GD.dword[0], ED->dword[0]); \
+            pthread_mutex_unlock(&emu->context->mutex_lock);\
+            break;                              \
+        case B+4: \
+            pthread_mutex_lock(&emu->context->mutex_lock);\
+            R_AL = OP##8(emu, R_AL, F8); \
+            pthread_mutex_unlock(&emu->context->mutex_lock);\
+            break;                              \
+        case B+5: \
+            pthread_mutex_lock(&emu->context->mutex_lock);\
+            R_EAX = OP##32(emu, R_EAX, F32); \
+            pthread_mutex_unlock(&emu->context->mutex_lock);\
+            break;
+#endif
+        GO(0x00, add)                   /* ADD 0x00 -> 0x05 */
+        GO(0x08, or)                    /*  OR 0x08 -> 0x0D */
+        GO(0x10, adc)                   /* ADC 0x10 -> 0x15 */
+        GO(0x18, sbb)                   /* SBB 0x18 -> 0x1D */
+        GO(0x20, and)                   /* AND 0x20 -> 0x25 */
+        GO(0x28, sub)                   /* SUB 0x28 -> 0x2D */
+        GO(0x30, xor)                   /* XOR 0x30 -> 0x35 */
+        #undef GO
+
+        case 0x0f:
+            opcode = F8;
+            switch (opcode) { 
+                case 0xB0:                      /* CMPXCHG Eb,Gb */
+                    CHECK_FLAGS(emu);
+                    nextop = F8;
+                    GET_EB;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                    do {
+                        tmp8u = arm_lock_read_b(EB);
+                        cmp8(emu, R_AL, tmp8u);
+                        if(ACCESS_FLAG(F_ZF)) {
+                            tmp32s = arm_lock_write_b(EB, GB);
+                        } else {
+                            R_AL = tmp8u;
+                            tmp32s = 0;
+                        }
+                    } while(tmp32s);
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    cmp8(emu, R_AL, EB->byte[0]);
+                    if(ACCESS_FLAG(F_ZF)) {
+                        EB->byte[0] = GB;
+                    } else {
+                        R_AL = EB->byte[0];
+                    }
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 0xB1:                      /* CMPXCHG Ed,Gd */
+                    nextop = F8;
+                    GET_ED;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                    if(((uintptr_t)ED)&3) {
+                        do {
+                            tmp32u = ED->dword[0] & ~0xff;
+                            tmp32u |= arm_lock_read_b(ED);
+                            cmp32(emu, R_EAX, tmp32u);
+                            if(ACCESS_FLAG(F_ZF)) {
+                                tmp32s = arm_lock_write_b(ED, GD.dword[0] & 0xff);
+                                if(!tmp32s)
+                                    ED->dword[0] = GD.dword[0];
+                            } else {
+                                R_EAX = tmp32u;
+                                tmp32s = 0;
+                            }
+                        } while(tmp32s);
+                    } else {
+                        do {
+                            tmp32u = arm_lock_read_d(ED);
+                            cmp32(emu, R_EAX, tmp32u);
+                            if(ACCESS_FLAG(F_ZF)) {
+                                tmp32s = arm_lock_write_d(ED, GD.dword[0]);
+                            } else {
+                                R_EAX = tmp32u;
+                                tmp32s = 0;
+                            }
+                        } while(tmp32s);
+                    }
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    cmp32(emu, R_EAX, ED->dword[0]);
+                    if(ACCESS_FLAG(F_ZF)) {
+                        ED->dword[0] = GD.dword[0];
+                    } else {
+                        R_EAX = ED->dword[0];
+                    }
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 0xB3:                      /* BTR Ed,Gd */
+                    CHECK_FLAGS(emu);
+                    nextop = F8;
+                    GET_ED;
+                    tmp8u = GD.byte[0];
+                    if((nextop&0xC0)!=0xC0)
+                    {
+                        #ifdef TEST_INTERPRETER
+                        test->memaddr=((test->memaddr)+(tmp8u>>5));
+                        *(uint32_t*)test->mem = *(uint32_t*)test->memaddr;
+                        #else
+                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                        #endif
+                    }
+                    tmp8u&=31;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                    do {
+                        tmp32u = arm_lock_read_d(ED);
+                        if(tmp32u & (1<<tmp8u)) {
+                            SET_FLAG(F_CF);
+                            tmp32u ^= (1<<tmp8u);
+                            tmp32s = arm_lock_write_d(ED, tmp32u);
+                        } else {
+                            CLEAR_FLAG(F_CF);
+                            tmp32s = 0;
+                        }
+                    } while(tmp32s);
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    if(ED->dword[0] & (1<<tmp8u)) {
+                        SET_FLAG(F_CF);
+                        ED->dword[0] ^= (1<<tmp8u);
+                    } else
+                        CLEAR_FLAG(F_CF);
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 0xBA:                      
+                    nextop = F8;
+                    switch((nextop>>3)&7) {
+                        case 4:                 /* BT Ed,Ib */
+                            CHECK_FLAGS(emu);
+                            GET_ED;
+                            tmp8u = F8;
+                            if((nextop&0xC0)!=0xC0)
+                            {
+                                #ifdef TEST_INTERPRETER
+                                test->memaddr=((test->memaddr)+(tmp8u>>5));
+                                *(uint32_t*)test->mem = *(uint32_t*)test->memaddr;
+                                #else
+                                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                                #endif
+                            }
+                            tmp8u&=31;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                            if(arm_lock_read_d(ED) & (1<<tmp8u))
+                                SET_FLAG(F_CF);
+                            else
+                                CLEAR_FLAG(F_CF);
+#else
+                            pthread_mutex_lock(&emu->context->mutex_lock);
+                            if(ED->dword[0] & (1<<tmp8u))
+                                SET_FLAG(F_CF);
+                            else
+                                CLEAR_FLAG(F_CF);
+                            pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                            break;
+                        case 5:             /* BTS Ed, Ib */
+                            CHECK_FLAGS(emu);
+                            GET_ED;
+                            tmp8u = F8;
+                            if((nextop&0xC0)!=0xC0)
+                            {
+                                #ifdef TEST_INTERPRETER
+                                test->memaddr=((test->memaddr)+(tmp8u>>5));
+                                *(uint32_t*)test->mem = *(uint32_t*)test->memaddr;
+                                #else
+                                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                                #endif
+                            }
+                            tmp8u&=31;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                            do {
+                                tmp32u = arm_lock_read_d(ED);
+                                if(tmp32u & (1<<tmp8u)) {
+                                    SET_FLAG(F_CF);
+                                    tmp32s = 0;
+                                } else {
+                                    CLEAR_FLAG(F_CF);
+                                    tmp32u ^= (1<<tmp8u);
+                                    tmp32s = arm_lock_write_d(ED, tmp32u);
+                                }
+                            } while(tmp32s);
+#else
+                            pthread_mutex_lock(&emu->context->mutex_lock);
+                            if(ED->dword[0] & (1<<tmp8u)) {
+                                SET_FLAG(F_CF);
+                            } else {
+                                CLEAR_FLAG(F_CF);
+                                ED->dword[0] ^= (1<<tmp8u);
+                            }
+                            pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                            break;
+                        case 6:             /* BTR Ed, Ib */
+                            CHECK_FLAGS(emu);
+                            GET_ED;
+                            tmp8u = F8;
+                            if((nextop&0xC0)!=0xC0)
+                            {
+                                #ifdef TEST_INTERPRETER
+                                test->memaddr=((test->memaddr)+(tmp8u>>5));
+                                *(uint32_t*)test->mem = *(uint32_t*)test->memaddr;
+                                #else
+                                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                                #endif
+                            }
+                            tmp8u&=31;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                            do {
+                                tmp32u = arm_lock_read_d(ED);
+                                if(tmp32u & (1<<tmp8u)) {
+                                    SET_FLAG(F_CF);
+                                    tmp32u ^= (1<<tmp8u);
+                                    tmp32s = arm_lock_write_d(ED, tmp32u);
+                                } else {
+                                    CLEAR_FLAG(F_CF);
+                                    tmp32s = 0;
+                                }
+                            } while(tmp32s);
+#else
+                            pthread_mutex_lock(&emu->context->mutex_lock);
+                            if(ED->dword[0] & (1<<tmp8u)) {
+                                SET_FLAG(F_CF);
+                                ED->dword[0] ^= (1<<tmp8u);
+                            } else
+                                CLEAR_FLAG(F_CF);
+                            pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                            break;
+
+                        default:
+                            return addr-3; //unfetchall 0F BA nextop but not F0, continue normal without LOCK
+                    }
+                    break;
+                case 0xBB:                      /* BTC Ed,Gd */
+                    CHECK_FLAGS(emu);
+                    nextop = F8;
+                    GET_ED;
+                    tmp8u = GD.byte[0];
+                    if((nextop&0xC0)!=0xC0)
+                    {
+                        #ifdef TEST_INTERPRETER
+                        test->memaddr=((test->memaddr)+(tmp8u>>5));
+                        *(uint32_t*)test->mem = *(uint32_t*)test->memaddr;
+                        #else
+                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                        #endif
+                    }
+                    tmp8u&=31;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                    do {
+                        tmp32u = arm_lock_read_d(ED);
+                        if(tmp32u & (1<<tmp8u)) {
+                            SET_FLAG(F_CF);
+                        } else {
+                            CLEAR_FLAG(F_CF);
+                        }
+                        tmp32u ^= (1<<tmp8u);
+                    } while(arm_lock_write_d(ED, tmp32u));
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    if(ED->dword[0] & (1<<tmp8u))
+                        SET_FLAG(F_CF);
+                    else
+                        CLEAR_FLAG(F_CF);
+                    ED->dword[0] ^= (1<<tmp8u);
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 0xC0:                      /* XADD Gb,Eb */
+                    nextop = F8;
+                    GET_EB;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                    do {
+                        tmp8u = arm_lock_read_b(EB);
+                        tmp8u2 = add8(emu, tmp8u, GB);
+                    } while (arm_lock_write_b(EB, tmp8u2));
+                    GB = tmp8u;
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    tmp8u = add8(emu, EB->byte[0], GB);
+                    GB = EB->byte[0];
+                    EB->byte[0] = tmp8u;
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 0xC1:                      /* XADD Gd,Ed */
+                    nextop = F8;
+                    GET_ED;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                    if(((uintptr_t)ED)&3) {
+                        do {
+                            tmp32u = ED->dword[0] & ~0xff;
+                            tmp32u |= arm_lock_read_b(ED);
+                            tmp32u2 = add32(emu, tmp32u, GD.dword[0]);
+                        } while(arm_lock_write_b(ED, tmp32u2&0xff));
+                        ED->dword[0] = tmp32u2;
+                    } else {
+                        do {
+                            tmp32u = arm_lock_read_d(ED);
+                            tmp32u2 = add32(emu, tmp32u, GD.dword[0]);
+                        } while(arm_lock_write_d(ED, tmp32u2));
+                    }
+                    GD.dword[0] = tmp32u;
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    tmp32u = add32(emu, ED->dword[0], GD.dword[0]);
+                    GD.dword[0] = ED->dword[0];
+                    ED->dword[0] = tmp32u;
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 0xC7:                      /* CMPXCHG8B Gq */
+                    nextop = F8;
+                    GET_ED8;
+                    switch((nextop>>3)&7) {
+                        case 1:
+                            CHECK_FLAGS(emu);
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                            if(((uintptr_t)ED)&7) {
+                                // unaligned!!!
+                                uint64_t ref = R_EDX;
+                                ref = ref<<32 | R_EAX;
+                                void* p = (void*)(((uintptr_t)ED + 3)&~3);
+                                int d = (((uintptr_t)p - (uintptr_t)ED))*8;
+                                do {
+                                    uint64_t m = *(uint64_t*)ED;
+                                    m&=~((0xffffffffLL)<<d);
+                                    tmp32u = arm_lock_read_d(p);
+                                    m|=((uint64_t)tmp32u)<<d;
+                                    if(ref == m) {
+                                        SET_FLAG(F_ZF);
+                                        m = ((uint64_t)R_ECX)<<32 | R_EBX;
+                                        tmp32u2 = m>>d;
+                                        tmp32s = arm_lock_write_d(p, tmp32u2);
+                                        if(!tmp32s)
+                                            *(uint64_t*)ED = m;
+                                    } else {
+                                        CLEAR_FLAG(F_ZF);
+                                        R_EAX = m&0xffffffff;
+                                        R_EDX = m>>32;
+                                        tmp32s = 0;
+                                    }
+                                } while(tmp32s);
+                            } else
+                            do {
+                                arm_lock_read_dd(&tmp32u, &tmp32u2, ED);
+                                if(R_EAX == tmp32u && R_EDX == tmp32u2) {
+                                    SET_FLAG(F_ZF);
+                                    tmp32s = arm_lock_write_dd(R_EBX, R_ECX, ED);
+                                } else {
+                                    CLEAR_FLAG(F_ZF);
+                                    R_EAX = tmp32u;
+                                    R_EDX = tmp32u2;
+                                    tmp32s = 0;
+                                }
+                            } while(tmp32s);
+#else
+                            pthread_mutex_lock(&emu->context->mutex_lock);
+                            tmp32u = ED->dword[0];
+                            tmp32u2= ED->dword[1];
+                            if(R_EAX == tmp32u && R_EDX == tmp32u2) {
+                                SET_FLAG(F_ZF);
+                                ED->dword[0] = R_EBX;
+                                ED->dword[1] = R_ECX;
+                            } else {
+                                CLEAR_FLAG(F_ZF);
+                                R_EAX = tmp32u;
+                                R_EDX = tmp32u2;
+                            }
+                            pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                            break;
+                        default:
+                            printf_log(LOG_NONE, "Illegal Opcode 0xF0 0xC7 0x%02X 0x%02X\n", nextop, PK(0));
+                                emu->quit=1;
+                                emu->error |= ERR_ILLEGAL;
+                            return 0;
+                    }
+                    break;
+                default:
+                    // trigger invalid lock?
+                    return addr-2; // unfetch 0F and nextop but not F0
+            }
+            break;
+
+        case 0x66:
+            #ifdef TEST_INTERPRETER
+            return TestF066(test, addr);
+            #else
+            return RunF066(emu, addr);
+            #endif
+
+        case 0x81:              /* GRP Ed,Id */
+        case 0x83:              /* GRP Ed,Ib */
+            nextop = F8;
+            GET_ED;
+            if(opcode==0x83) {
+                tmp32s = F8S;
+                tmp32u = (uint32_t)tmp32s;
+            } else
+                tmp32u = F32;
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+            switch((nextop>>3)&7) {
+                case 0: do { tmp32u2 = arm_lock_read_d(ED);} while(arm_lock_write_d(ED, add32(emu, tmp32u2, tmp32u))); break;
+                case 1: do { tmp32u2 = arm_lock_read_d(ED);} while(arm_lock_write_d(ED,  or32(emu, tmp32u2, tmp32u))); break;
+                case 2: do { tmp32u2 = arm_lock_read_d(ED);} while(arm_lock_write_d(ED, adc32(emu, tmp32u2, tmp32u))); break;
+                case 3: do { tmp32u2 = arm_lock_read_d(ED);} while(arm_lock_write_d(ED, sbb32(emu, tmp32u2, tmp32u))); break;
+                case 4: do { tmp32u2 = arm_lock_read_d(ED);} while(arm_lock_write_d(ED, and32(emu, tmp32u2, tmp32u))); break;
+                case 5: do { tmp32u2 = arm_lock_read_d(ED);} while(arm_lock_write_d(ED, sub32(emu, tmp32u2, tmp32u))); break;
+                case 6: do { tmp32u2 = arm_lock_read_d(ED);} while(arm_lock_write_d(ED, xor32(emu, tmp32u2, tmp32u))); break;
+                case 7:                cmp32(emu, ED->dword[0], tmp32u); break;
+            }
+#else
+            pthread_mutex_lock(&emu->context->mutex_lock);
+            switch((nextop>>3)&7) {
+                case 0: ED->dword[0] = add32(emu, ED->dword[0], tmp32u); break;
+                case 1: ED->dword[0] =  or32(emu, ED->dword[0], tmp32u); break;
+                case 2: ED->dword[0] = adc32(emu, ED->dword[0], tmp32u); break;
+                case 3: ED->dword[0] = sbb32(emu, ED->dword[0], tmp32u); break;
+                case 4: ED->dword[0] = and32(emu, ED->dword[0], tmp32u); break;
+                case 5: ED->dword[0] = sub32(emu, ED->dword[0], tmp32u); break;
+                case 6: ED->dword[0] = xor32(emu, ED->dword[0], tmp32u); break;
+                case 7:                cmp32(emu, ED->dword[0], tmp32u); break;
+            }
+            pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+            break;
+        case 0x86:                      /* XCHG Eb,Gb */
+        case 0x87:                      /* XCHG Ed,Gd */
+            return addr-1;   // let the normal XCHG execute, it have integrated LOCK
+        case 0x90:  // This is xchg reg, reg, so let's ignore the LOCK prefix...
+        case 0x91:
+        case 0x92:
+        case 0x93:
+        case 0x94:
+        case 0x95:
+        case 0x96:
+        case 0x97:
+            return addr-1;
+
+        case 0xFF:              /* GRP 5 Ed */
+            nextop = F8;
+            GET_ED;
+            switch((nextop>>3)&7) {
+                case 0:                 /* INC Ed */
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                    if((uintptr_t)ED&3) { 
+                        //meh.
+                        do {
+                            tmp32u = ED->dword[0];
+                            tmp32u &=~0xff;
+                            tmp32u |= arm_lock_read_b(ED);
+                            tmp32u = inc32(emu, tmp32u);
+                        } while(arm_lock_write_b(ED, tmp32u&0xff));
+                        ED->dword[0] = tmp32u;
+                    } else
+                        do {
+                            tmp32u = arm_lock_read_d(ED);
+                        } while(arm_lock_write_d(ED, inc32(emu, tmp32u)));
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    ED->dword[0] = inc32(emu, ED->dword[0]);
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 1:                 /* DEC Ed */
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+                    if((uintptr_t)ED&3) { 
+                        //meh.
+                        do {
+                            tmp32u = ED->dword[0];
+                            tmp32u &=~0xff;
+                            tmp32u |= arm_lock_read_b(ED);
+                            tmp32u = dec32(emu, tmp32u);
+                        } while(arm_lock_write_b(ED, tmp32u&0xff));
+                        ED->dword[0] = tmp32u;
+                    } else
+                        do {
+                            tmp32u = arm_lock_read_d(ED);
+                        } while(arm_lock_write_d(ED, dec32(emu, tmp32u)));
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    ED->dword[0] = dec32(emu, ED->dword[0]);
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                default:
+                    printf_log(LOG_NONE, "Illegal Opcode 0xF0 0xFF 0x%02X 0x%02X\n", nextop, PK(0));
+                    emu->quit=1;
+                    emu->error |= ERR_ILLEGAL;
+                    return 0;
+            }
+            break;
+
+        default:
+            return 0;
+    }
+    return addr;
+}

src/emu/x86runf066.c

@@ -0,0 +1,184 @@
+#define _GNU_SOURCE
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "x87emu_private.h"
+#include "box86context.h"
+#include "my_cpuid.h"
+#include "bridge.h"
+#include "signals.h"
+#if defined(DYNAREC) && !defined(TEST_INTERPRETER)
+#include "../dynarec/arm_lock_helper.h"
+#endif
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestF066(x86test_t *test, uintptr_t addr)
+#else
+uintptr_t RunF066(x86emu_t *emu, uintptr_t addr)
+#endif
+{
+    uint8_t opcode;
+    uint8_t nextop;
+    reg32_t *oped;
+    uint8_t tmp8u, tmp8u2;
+    int8_t tmp8s;
+    uint16_t tmp16u, tmp16u2;
+    int16_t tmp16s;
+    uint32_t tmp32u, tmp32u2, tmp32u3;
+    int32_t tmp32s, tmp32s2;
+    uint64_t tmp64u;
+    int64_t tmp64s;
+    double d;
+    float f;
+    int64_t ll;
+    sse_regs_t *opex, eax1;
+    mmx87_regs_t *opem, eam1;
+
+#ifdef TEST_INTERPRETER
+    x86emu_t *emu = test->emu;
+#endif
+    opcode = F8;
+
+    switch(opcode) {
+
+        case 0x0f:
+            opcode = F8;
+            switch (opcode) { 
+                case 0xB1:                      /* CMPXCHG Ew,Gw */
+                    nextop = F8;
+                    GET_EW;
+#ifdef DYNAREC
+                    do {
+                        tmp16u = arm_lock_read_h(ED);
+                        cmp16(emu, R_AX, tmp16u);
+                        if(ACCESS_FLAG(F_ZF)) {
+                            tmp32s = arm_lock_write_h(ED, GW.word[0]);
+                        } else {
+                            R_AX = tmp16u;
+                            tmp32s = 0;
+                        }
+                    } while(tmp32s);
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    GET_EW;
+                    cmp16(emu, R_AX, EW->word[0]);
+                    if(ACCESS_FLAG(F_ZF)) {
+                        EW->word[0] = GW.word[0];
+                    } else {
+                        R_AX = EW->word[0];
+                    }
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+            default:
+                return 0;
+            }
+            break;
+        case 0x81:              /* GRP Ed,Iw */
+        case 0x83:              /* GRP Ed,Ib */
+            nextop = F8;
+            GET_EW;
+            if(opcode==0x83) {
+                tmp16s = F8S;
+                tmp16u = (uint32_t)tmp16s;
+            } else
+                tmp16u = F16;
+#ifdef DYNAREC
+            switch((nextop>>3)&7) {
+                case 0: do { tmp16u2 = arm_lock_read_h(EW);} while(arm_lock_write_h(EW, add16(emu, tmp16u2, tmp16u))); break;
+                case 1: do { tmp16u2 = arm_lock_read_h(EW);} while(arm_lock_write_h(EW,  or16(emu, tmp16u2, tmp16u))); break;
+                case 2: do { tmp16u2 = arm_lock_read_h(EW);} while(arm_lock_write_h(EW, adc16(emu, tmp16u2, tmp16u))); break;
+                case 3: do { tmp16u2 = arm_lock_read_h(EW);} while(arm_lock_write_h(EW, sbb16(emu, tmp16u2, tmp16u))); break;
+                case 4: do { tmp16u2 = arm_lock_read_h(EW);} while(arm_lock_write_h(EW, and16(emu, tmp16u2, tmp16u))); break;
+                case 5: do { tmp16u2 = arm_lock_read_h(EW);} while(arm_lock_write_h(EW, sub16(emu, tmp16u2, tmp16u))); break;
+                case 6: do { tmp16u2 = arm_lock_read_h(EW);} while(arm_lock_write_h(EW, xor16(emu, tmp16u2, tmp16u))); break;
+                case 7:                cmp16(emu, ED->dword[0], tmp16u); break;
+            }
+#else
+            pthread_mutex_lock(&emu->context->mutex_lock);
+            switch((nextop>>3)&7) {
+                case 0: EW->word[0] = add16(emu, EW->word[0], tmp16u); break;
+                case 1: EW->word[0] =  or16(emu, EW->word[0], tmp16u); break;
+                case 2: EW->word[0] = adc16(emu, EW->word[0], tmp16u); break;
+                case 3: EW->word[0] = sbb16(emu, EW->word[0], tmp16u); break;
+                case 4: EW->word[0] = and16(emu, EW->word[0], tmp16u); break;
+                case 5: EW->word[0] = sub16(emu, EW->word[0], tmp16u); break;
+                case 6: EW->word[0] = xor16(emu, EW->word[0], tmp16u); break;
+                case 7:               cmp16(emu, EW->word[0], tmp16u); break;
+            }
+            pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+            break;
+        
+        case 0xFF:
+            nextop = F8;
+            GET_EW;
+            switch((nextop>>3)&7) {
+                case 0:                 /* INC Ed */
+#ifdef DYNAREC
+                    if((uintptr_t)ED&1) { 
+                        //meh.
+                        do {
+                            tmp16u = ED->dword[0];
+                            tmp16u &=~0xff;
+                            tmp16u |= arm_lock_read_b(ED);
+                            tmp16u = inc16(emu, tmp16u);
+                        } while(arm_lock_write_b(ED, tmp16u&0xff));
+                        ED->dword[0] = tmp16u;
+                    } else
+                        do {
+                            tmp16u = arm_lock_read_h(ED);
+                        } while(arm_lock_write_h(ED, inc16(emu, tmp16u)));
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    ED->dword[0] = inc32(emu, ED->dword[0]);
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 1:                 /* DEC Ed */
+#ifdef DYNAREC
+                    if((uintptr_t)ED&1) { 
+                        //meh.
+                        do {
+                            tmp16u = ED->dword[0];
+                            tmp16u &=~0xff;
+                            tmp16u |= arm_lock_read_b(ED);
+                            tmp16u = dec16(emu, tmp16u);
+                        } while(arm_lock_write_b(ED, tmp16u&0xff));
+                        ED->dword[0] = tmp16u;
+                    } else
+                        do {
+                            tmp16u = arm_lock_read_h(ED);
+                        } while(arm_lock_write_h(ED, dec16(emu, tmp16u)));
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    ED->dword[0] = dec32(emu, ED->dword[0]);
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                default:
+                    return 0;
+            }
+            break;
+
+        default:
+            return 0;
+    }
+    return addr;
+}

src/emu/x86runf20f.c

@@ -1,4 +1,51 @@
+#define _GNU_SOURCE
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "debug.h"
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run.h"
+#include "x86emu_private.h"
+#include "x86run_private.h"
+#include "x86primop.h"
+#include "x86trace.h"
+#include "x87emu_private.h"
+#include "box86context.h"
+#include "bridge.h"
+
+#include "modrm.h"
+
+#ifdef TEST_INTERPRETER
+uintptr_t TestF20F(x86test_t *test, uintptr_t addr, int *step)
+#else
+uintptr_t RunF20F(x86emu_t *emu, uintptr_t addr, int *step)
+#endif
+{
+    uint8_t opcode;
+    uint8_t nextop;
+    int8_t tmp8s;
+    uint8_t tmp8u;
+    int32_t tmp32s;
+    int64_t tmp64s0, tmp64s1;
+    reg32_t *oped, *opgd;
+    sse_regs_t *opex, *opgx, eax1;
+    mmx87_regs_t *opgm;
+    #ifndef NOALIGN
+    int is_nan;
+    #endif
+    #ifdef TEST_INTERPRETER
+    x86emu_t*emu = test->emu;
+    #endif
+
     opcode = F8;
+
     switch(opcode) {
 
     case 0x10:  /* MOVSD Gx, Ex */
@@ -30,25 +77,31 @@
     case 0x2C:  /* CVTTSD2SI Gd, Ex */
         nextop = F8;
         GET_EX;
-        GD.sdword[0] = EX->d[0];
+        if(!isfinite(EX->d[0]) || EX->d[0]>(double)0x7fffffff || EX->d[0]<-(double)0x80000000U)
+            GD.dword[0] = 0x80000000;
+        else
+            GD.sdword[0] = EX->d[0];
         break;
     case 0x2D:  /* CVTSD2SI Gd, Ex */
         nextop = F8;
         GET_EX;
-        switch((emu->mxcsr>>13)&3) {
-            case ROUND_Nearest:
-                GD.sdword[0] = floor(EX->d[0]+0.5);
-                break;
-            case ROUND_Down:
-                GD.sdword[0] = floor(EX->d[0]);
-                break;
-            case ROUND_Up:
-                GD.sdword[0] = ceil(EX->d[0]);
-                break;
-            case ROUND_Chop:
-                GD.sdword[0] = EX->d[0];
-                break;
-        }
+        if(!isfinite(EX->d[0]) || EX->d[0]>(double)0x7fffffff || EX->d[0]<-(double)0x80000000U)
+            GD.dword[0] = 0x80000000;
+        else
+            switch(emu->mxcsr.f.MXCSR_RC) {
+                case ROUND_Nearest:
+                    GD.sdword[0] = nearbyint(EX->d[0]);
+                    break;
+                case ROUND_Down:
+                    GD.sdword[0] = floor(EX->d[0]);
+                    break;
+                case ROUND_Up:
+                    GD.sdword[0] = ceil(EX->d[0]);
+                    break;
+                case ROUND_Chop:
+                    GD.sdword[0] = EX->d[0];
+                    break;
+            }
         break;
     case 0x51:  /* SQRTSD Gx, Ex */
         nextop = F8;
@@ -119,7 +172,7 @@
     case 0x5F:  /* MAXSD Gx, Ex */
         nextop = F8;
         GET_EX;
-        if (isnan(GX.d[0]) || isnan(EX->d[0]) || isgreater(EX->d[0], GX.d[0]))
+        if (isnan(GX.d[0]) || isnan(EX->d[0]) || isgreaterequal(EX->d[0], GX.d[0]))
             GX.d[0] = EX->d[0];
         break;
 
@@ -151,6 +204,19 @@
             GX.f[3] = EX->f[2] + EX->f[3];
         }
         break;
+    case 0x7D:  /* HSUBPS Gx, Ex */
+        nextop = F8;
+        GET_EX;
+        GX.f[0] -= GX.f[1];
+        GX.f[1] = GX.f[2] - GX.f[3];
+        if(EX==&GX) {
+            GX.f[2] = GX.f[0];
+            GX.f[3] = GX.f[1];
+        } else {
+            GX.f[2] = EX->f[0] - EX->f[1];
+            GX.f[3] = EX->f[2] - EX->f[3];
+        }
+        break;
 
     #define GOCOND(BASE, PREFIX, CONDITIONAL)   \
     case BASE+0:                                \
@@ -236,7 +302,7 @@
         
     GOCOND(0x80
         , tmp32s = F32S; CHECK_FLAGS(emu);
-        , ip += tmp32s;
+        , addr += tmp32s; STEP3;
     )                               /* 0x80 -> 0x8F Jxx */
         
     case 0xC2:  /* CMPSD Gx, Ex, Ib */
@@ -275,24 +341,24 @@
     case 0xE6:  /* CVTPD2DQ Gx, Ex */
         nextop = F8;
         GET_EX;
-        switch((emu->mxcsr>>13)&3) {
-            case ROUND_Nearest:
-                GX.sd[0] = floor(EX->d[0]+0.5);
-                GX.sd[1] = floor(EX->d[1]+0.5);
-                break;
-            case ROUND_Down:
-                GX.sd[0] = floor(EX->d[0]);
-                GX.sd[1] = floor(EX->d[1]);
-                break;
-            case ROUND_Up:
-                GX.sd[0] = ceil(EX->d[0]);
-                GX.sd[1] = ceil(EX->d[1]);
-                break;
-            case ROUND_Chop:
-                GX.sd[0] = EX->d[0];
-                GX.sd[1] = EX->d[1];
-                break;
-        }
+        for(int i=0; i<2; ++i)
+            if(!isfinite(EX->d[i]) || EX->d[i]>(double)0x7fffffff || EX->d[i]<-(double)0x80000000U)
+                GX.ud[i] = 0x80000000;
+            else
+                switch(emu->mxcsr.f.MXCSR_RC) {
+                    case ROUND_Nearest:
+                        GX.sd[i] = nearbyint(EX->d[i]);
+                        break;
+                    case ROUND_Down:
+                        GX.sd[i] = floor(EX->d[i]);
+                        break;
+                    case ROUND_Up:
+                        GX.sd[i] = ceil(EX->d[i]);
+                        break;
+                    case ROUND_Chop:
+                        GX.sd[i] = EX->d[i];
+                        break;
+                }
         GX.q[1] = 0;
         break;
 
@@ -303,6 +369,7 @@
         break;
 
     default:
-        goto _default;
+        return 0;
     }
-
+    return addr;
+}