Docs: Update for release FEX-2207

AOTIR: Fix IRList delete

.github/workflows/ccpp.yml

@@ -29,6 +29,20 @@ jobs:
     - name: Set runner label
       run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
 
+    - name: Set rootfs paths
+      run: |
+        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+
+    - name: Update RootFS cache
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: $GITHUB_WORKSPACE/Scripts/CI_FetchRootFS.py
+
     - name : submodule checkout
       # Need to update submodules
       run: |
@@ -51,7 +65,7 @@ jobs:
       # Note the current convention is to use the -S and -B options here to specify source
       # and build directories, but this is only available with CMake 3.13 and higher.
       # The CMake binaries on the Github Actions machines are (as of this writing) 3.12
-      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DENABLE_INTERPRETER=True
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DENABLE_INTERPRETER=True -DBUILD_FEX_LINUX_TESTS=True
 
     - name: Build
       working-directory: ${{runner.workspace}}/build
@@ -153,6 +167,17 @@ jobs:
       working-directory: ${{runner.workspace}}/build
       run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_APITests.log || true
 
+    - name: FEXLinuxTests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target fex_linux_tests_all
+
+    - name: FEXLinuxTests Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_FEXLinuxTests.log || true
+
     - name: Truncate test results
       if: ${{ always() }}
       shell: bash

CMakeLists.txt

@@ -1,16 +1,22 @@
 cmake_minimum_required(VERSION 3.14)
 project(FEX)
 
+INCLUDE (CheckIncludeFiles)
+CHECK_INCLUDE_FILES ("gdb/jit-reader.h" HAVE_GDB_JIT_READER_H)
+
 option(BUILD_TESTS "Build unit tests to ensure sanity" TRUE)
+option(BUILD_FEX_LINUX_TESTS "Build FEXLinuxTests, requires x86 compiler" FALSE)
 option(BUILD_THUNKS "Build thunks" FALSE)
 option(ENABLE_CLANG_FORMAT "Run clang format over the source" FALSE)
 option(ENABLE_IWYU "Enables include what you use program" FALSE)
 option(ENABLE_LTO "Enable LTO with compilation" TRUE)
 option(ENABLE_XRAY "Enable building with LLVM X-Ray" FALSE)
-option(ENABLE_LLD "Enable linking with LLD" FALSE)
+option(ENABLE_LLD "Enable linking with lld" FALSE)
+option(ENABLE_MOLD "Enable linking with mold" FALSE)
 option(ENABLE_ASAN "Enables Clang ASAN" FALSE)
 option(ENABLE_TSAN "Enables Clang TSAN" FALSE)
 option(ENABLE_ASSERTIONS "Enables assertions in build" FALSE)
+option(ENABLE_GDB_SYMBOLS "Enables GDBSymbols integration support" ${HAVE_GDB_JIT_READER_H})
 option(ENABLE_VISUAL_DEBUGGER "Enables the visual debugger for compiling" FALSE)
 option(ENABLE_STRICT_WERROR "Enables stricter -Werror for CI" FALSE)
 option(ENABLE_WERROR "Enables -Werror" FALSE)
@@ -21,9 +27,9 @@ option(ENABLE_COMPILE_TIME_TRACE "Enables time trace compile option" FALSE)
 option(ENABLE_LIBCXX "Enables LLVM libc++" FALSE)
 option(ENABLE_INTERPRETER "Enables FEX's Interpreter" FALSE)
 option(ENABLE_CCACHE "Enables ccache for compile caching" TRUE)
+option(ENABLE_TERMUX_BUILD "Forces building for Termux on a non-Termux build machine" FALSE)
 
-set (X86_C_COMPILER "x86_64-linux-gnu-gcc" CACHE STRING "c compiler for compiling x86 guest libs")
-set (X86_CXX_COMPILER "x86_64-linux-gnu-g++" CACHE STRING "c++ compiler for compiling x86 guest libs")
+set (X86_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/toolchain_x86.cmake" CACHE FILEPATH "Toolchain file for the x86 (cross-)compiler")
 set (DATA_DIRECTORY "${CMAKE_INSTALL_PREFIX}/share/fex-emu" CACHE PATH "global data directory")
 
 # These options are meant for package management
@@ -41,6 +47,12 @@ if (ENABLE_ASSERTIONS)
   add_definitions(-DASSERTIONS_ENABLED=1)
 endif()
 
+if (ENABLE_GDB_SYMBOLS)
+  message(STATUS "GDBSymbols support enabled")
+  add_definitions(-DGDB_SYMBOLS_ENABLED=1)
+endif()
+
+
 if (ENABLE_INTERPRETER)
   message(STATUS "Interpreter enabled")
   add_definitions(-DINTERPRETER_ENABLED=1)
@@ -73,6 +85,7 @@ if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
   set(_M_X86_64 1)
   add_definitions(-D_M_X86_64=1)
   set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
+  set (X86_TOOLCHAIN_FILE "")
 endif()
 
 if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
@@ -99,9 +112,14 @@ if (ENABLE_COMPILE_TIME_TRACE)
 endif()
 
 set (PTHREAD_LIB pthread)
-if (ENABLE_LLD)
+
+if (ENABLE_LLD AND ENABLE_MOLD)
+  message (FATAL_ERROR "Cannot enable both lld and mold")
+elseif (ENABLE_LLD)
   set (LD_OVERRIDE "-fuse-ld=lld")
-  link_libraries(${LD_OVERRIDE})
+  add_link_options(${LD_OVERRIDE})
+elseif (ENABLE_MOLD)
+  add_link_options("-fuse-ld=mold")
 endif()
 
 if (ENABLE_LIBCXX)
@@ -115,59 +133,7 @@ if (NOT ENABLE_OFFLINE_TELEMETRY)
   add_definitions(-DFEX_DISABLE_TELEMETRY=1)
 endif()
 
-# Check if the build target page size is 4096
-include(CheckCSourceRuns)
-
-check_c_source_runs(
-  "#include <unistd.h>
-  int main(int argc, char* argv[])
-  {
-    return getpagesize() == 4096 ? 0 : 1;
-  }"
-  PAGEFILE_RESULT
-  )
-
-if (NOT ${PAGEFILE_RESULT})
-  message(FATAL_ERROR "Host PAGE_SIZE is not 4096. Can't build on this target")
-endif()
-
-include(CheckCXXSourceCompiles)
-check_cxx_source_compiles(
-"#include <sys/user.h>
- int main() {
-  return PAGE_SIZE;
- }
- "
- HAS_PAGESIZE)
-
-check_cxx_source_compiles(
-"#include <sys/user.h>
- int main() {
-  return PAGE_SHIFT;
- }
- "
- HAS_PAGESHIFT)
-
-check_cxx_source_compiles(
-"#include <sys/user.h>
- int main() {
-  return PAGE_MASK;
- }
- "
- HAS_PAGEMASK)
-
-if (NOT HAS_PAGESIZE)
-  add_definitions(-DPAGE_SIZE=4096)
-endif()
-
-if (NOT HAS_PAGESHIFT)
-  add_definitions(-DPAGE_SHIFT=12)
-endif()
-if (NOT HAS_PAGEMASK)
-  add_definitions("-DPAGE_MASK=(~(PAGE_SIZE-1))")
-endif()
-
-if(DEFINED ENV{TERMUX_VERSION})
+if(DEFINED ENV{TERMUX_VERSION} OR ENABLE_TERMUX_BUILD)
   add_definitions(-DTERMUX_BUILD=1)
   set(TERMUX_BUILD 1)
   # Termux doesn't support Jemalloc due to bad interactions between emutls, jemalloc, and scudo
@@ -176,7 +142,7 @@ endif()
 
 if (ENABLE_STATIC_PIE)
   if (_M_ARM_64 AND ENABLE_LLD)
-    message (FATAL_ERROR "Static linking does not currently work with AArch64+LLD. Use GNU ld for now.")
+    message (FATAL_ERROR "Static linking does not currently work with AArch64+lld. Use GNU ld for now.")
   endif()
 
   file(WRITE ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt.c
@@ -558,8 +524,7 @@ if (BUILD_THUNKS)
     BINARY_DIR "Guest"
     CMAKE_ARGS
       "-DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}"
-      "-DX86_C_COMPILER:STRING=${X86_C_COMPILER}"
-      "-DX86_CXX_COMPILER:STRING=${X86_CXX_COMPILER}"
+      "-DCMAKE_TOOLCHAIN_FILE:FILEPATH=${X86_TOOLCHAIN_FILE}"
       "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
       "-DSTRUCT_VERIFIER=${CMAKE_SOURCE_DIR}/Scripts/StructPackVerifier.py"
       "-DGENERATOR_EXE=$<TARGET_FILE:thunkgen>"

Data/AppConfig/steamwebhelper.json

@@ -0,0 +1,5 @@
+{
+  "Config": {
+    "AdditionalArguments": "--no-sandbox"
+  }
+}

Data/ThunksDB.json

@@ -6,18 +6,10 @@
         "X11"
       ],
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGL.so",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1.7.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1.7.0",
-        "/lib/x86_64-linux-gnu/libGL.so",
-        "/lib/x86_64-linux-gnu/libGL.so.1",
-        "/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/lib/x86_64-linux-gnu/libGL.so.1.7.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libGL.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libGL.so.1",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libGL.so.1.2.0",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libGL.so.1.7.0"
       ]
     },
     "GLESv2": {
@@ -26,322 +18,151 @@
         "X11"
       ],
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so.2.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so.2.0.0",
-        "/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/lib/x86_64-linux-gnu/libGLESv2.so.2.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libGLESv2.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libGLESv2.so.2",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libGLESv2.so.2.0.0"
       ]
     },
     "X11": {
       "Library": "libX11-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libX11.so",
-        "/usr/lib/x86_64-linux-gnu/libX11.so.6",
-        "/usr/lib/x86_64-linux-gnu/libX11.so.6.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so.6",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so.6.4.0",
-        "/lib/x86_64-linux-gnu/libX11.so",
-        "/lib/x86_64-linux-gnu/libX11.so.6",
-        "/lib/x86_64-linux-gnu/libX11.so.6.4.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libX11.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libX11.so.6",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libX11.so.6.4.0"
       ]
     },
-    "Vulkan-radeon": {
-      "Library": "libvulkan_radeon-guest.so",
+    "Vulkan": {
+      "Library": "libvulkan-guest.so",
       "Depends": [
         "xcb"
       ],
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_radeon.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_radeon.so",
-        "/lib/x86_64-linux-gnu/libvulkan_radeon.so"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libvulkan.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libvulkan.so.1",
+        "@HOME@/.local/share/Steam/ubuntu12_32/steam-runtime/pinned_libs_64/libvulkan.so.1"
       ],
       "Comment": [
         "Vulkan library relies on xcb, otherwise it crashes with jemalloc"
       ]
     },
-    "Vulkan-lavapipe": {
-      "Library": "libvulkan_lvp-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_lvp.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_lvp.so",
-        "/lib/x86_64-linux-gnu/libvulkan_lvp.so"
-      ]
-    },
-    "Vulkan-freedreno": {
-      "Library": "libvulkan_freedreno-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_freedreno.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_freedreno.so",
-        "/lib/x86_64-linux-gnu/libvulkan_freedreno.so"
-      ]
-    },
-    "Vulkan-intel": {
-      "Library": "libvulkan_intel-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_intel.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_intel.so",
-        "/lib/x86_64-linux-gnu/libvulkan_intel.so"
-      ]
-    },
-    "Vulkan-panfrost": {
-      "Library": "libvulkan_panfrost-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_panfrost.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_panfrost.so",
-        "/lib/x86_64-linux-gnu/libvulkan_panfrost.so"
-      ]
-    },
-    "Vulkan-nvidia": {
-      "Library": "libvulkan_nvidia-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGLX_nvidia.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGLX_nvidia.so.0",
-        "/lib/x86_64-linux-gnu/libGLX_nvidia.so.0"
-      ],
-      "Comment": [
-        "Not currently wired up"
-      ]
-    },
-    "Vulkan-virtio": {
-      "Library": "libvulkan_virtio-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_virtio.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_virtio.so",
-        "/lib/x86_64-linux-gnu/libvulkan_virtio.so"
-      ]
-    },
     "xcb": {
       "Library": "libxcb-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxcb.so.1.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so.1.1.0",
-        "/lib/x86_64-linux-gnu/libxcb.so",
-        "/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/lib/x86_64-linux-gnu/libxcb.so.1.1.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb.so.1",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb.so.1.1.0"
       ]
     },
     "xcb-dri2": {
       "Library": "libxcb_dri2-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri2.so.0.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri2.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri2.so.0",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri2.so.0.0.0"
       ]
     },
     "xcb-dri3": {
       "Library": "libxcb_dri3-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri3.so.0.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri3.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri3.so.0",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri3.so.0.0.0"
       ]
     },
     "xcb-xfixes": {
       "Library": "libxcb_xfixes-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-xfixes.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-xfixes.so.0",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0"
       ]
     },
     "xcb-shm": {
       "Library": "libxcb_shm-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-shm.so.0.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-shm.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-shm.so.0",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-shm.so.0.0.0"
       ]
     },
     "xcb-sync": {
       "Library": "libxcb_sync-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so.1.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so.1.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/lib/x86_64-linux-gnu/libxcb-sync.so.1.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-sync.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-sync.so.1",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-sync.so.1.0.0"
       ]
     },
     "xcb-randr": {
       "Library": "libxcb_randr-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so.0.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so.0.1.0",
-        "/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-randr.so.0.1.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-randr.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-randr.so.0",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-randr.so.0.1.0"
       ]
     },
     "xcb-present": {
       "Library": "libxcb_present-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-present.so.0.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-present.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-present.so.0",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-present.so.0.0.0"
       ]
     },
     "xcb-glx": {
       "Library": "libxcb_glx-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-glx.so.0.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-glx.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-glx.so.0",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-glx.so.0.0.0"
       ]
     },
     "xshmfence": {
       "Library": "libshmfence-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so.1.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so.1.0.0",
-        "/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/lib/x86_64-linux-gnu/libxshmfence.so.1.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxshmfence.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxshmfence.so.1",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libxshmfence.so.1.0.0"
       ]
     },
     "drm": {
       "Library": "libdrm-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libdrm.so",
-        "/usr/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/usr/lib/x86_64-linux-gnu/libdrm.so.2.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so.2.4.0",
-        "/lib/x86_64-linux-gnu/libdrm.so",
-        "/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/lib/x86_64-linux-gnu/libdrm.so.2.4.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libdrm.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libdrm.so.2",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libdrm.so.2.4.0"
       ]
     },
     "asound": {
       "Library": "libasound-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libasound.so",
-        "/usr/lib/x86_64-linux-gnu/libasound.so.2",
-        "/usr/lib/x86_64-linux-gnu/libasound.so.2.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so.2.0.0",
-        "/lib/x86_64-linux-gnu/libasound.so",
-        "/lib/x86_64-linux-gnu/libasound.so.2",
-        "/lib/x86_64-linux-gnu/libasound.so.2.0.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libasound.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libasound.so.2",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libasound.so.2.0.0"
       ]
     },
     "Xrender": {
       "Library": "libXrender-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libXrender.so",
-        "/usr/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/usr/lib/x86_64-linux-gnu/libXrender.so.1.3.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so.1.3.0",
-        "/lib/x86_64-linux-gnu/libXrender.so",
-        "/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/lib/x86_64-linux-gnu/libXrender.so.1.3.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXrender.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXrender.so.1",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXrender.so.1.3.0"
       ]
     },
     "Xext": {
       "Library": "libXext-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libXext.so",
-        "/usr/lib/x86_64-linux-gnu/libXext.so.6",
-        "/usr/lib/x86_64-linux-gnu/libXext.so.6.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so.6",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so.6.4.0",
-        "/lib/x86_64-linux-gnu/libXext.so",
-        "/lib/x86_64-linux-gnu/libXext.so.6",
-        "/lib/x86_64-linux-gnu/libXext.so.6.4.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXext.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXext.so.6",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXext.so.6.4.0"
       ]
     },
     "Xfixes": {
       "Library": "libXfixes-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so",
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so.3.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so.3.1.0",
-        "/lib/x86_64-linux-gnu/libXfixes.so",
-        "/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/lib/x86_64-linux-gnu/libXfixes.so.3.1.0"
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXfixes.so",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXfixes.so.3",
+        "@PREFIX_LIB@/x86_64-linux-gnu/libXfixes.so.3.1.0"
       ]
     },
     "":{}

External/FEXCore/Readme.md

@@ -18,14 +18,8 @@ This project aims to provide a fast and functional x86-64 emulation library that
 * Portable library implementation in order to support easy integration in to applications
 ### Target Host Architecture
 The target host architecture for this library is AArch64. Specifically the ARMv8.1 version or newer.
-The CPU IR is designed with AArch64 in mind but there is a desire to run the recompiled code on other architectures as well.
-Multiple architecture support is desired for easier bringup and debugging, performance isn't as much of a priority there (ex. x86-64(guest) translated to x86-64(host))
-### Not currently goals but will be in the future
-* 32bit x86 support
-  * This will be a desire in the future, but to lower the amount of work required, decided to push this off for now.
-* Integration in to WINE
-* Later generation of x86-64 instruction sets
-  * Including AVX, F16C, XOP, FMA, AVX2, etc
+The CPU IR is designed with AArch64 in mind but should allow for other architectures as well.
+x86-64 host support is available for ease of development, but is not a priority.
 ### Not desired
 * Kernel space emulation
 * CPL0-2 emulation
@@ -33,7 +27,3 @@ Multiple architecture support is desired for easier bringup and debugging, perfo
 * IRQs
 * SVM
 * "Cycle Accurate" emulation
-### Dependencies
- * clang-tidy if you want to ensure the code stays tidy
- * cmake
- * A C++17 compliant compiler (There are assumptions made about using Clang and LTO)

External/FEXCore/Source/CMakeLists.txt

@@ -80,16 +80,20 @@ set (SRCS
   Interface/Context/Context.cpp
   Interface/Core/LookupCache.cpp
   Interface/Core/BlockSamplingData.cpp
-  Interface/Core/CompileService.cpp
   Interface/Core/Core.cpp
+  Interface/Core/CPUBackend.cpp
   Interface/Core/CPUID.cpp
   Interface/Core/Frontend.cpp
   Interface/Core/GdbServer.cpp
   Interface/Core/HostFeatures.cpp
+  Interface/Core/ObjectCache/JobHandling.cpp
+  Interface/Core/ObjectCache/NamedRegionObjectHandler.cpp
+  Interface/Core/ObjectCache/ObjectCacheService.cpp
   Interface/Core/OpcodeDispatcher/Crypto.cpp
   Interface/Core/OpcodeDispatcher/Flags.cpp
   Interface/Core/OpcodeDispatcher/Vector.cpp
   Interface/Core/OpcodeDispatcher/X87.cpp
+  Interface/Core/OpcodeDispatcher/X87F64.cpp
   Interface/Core/OpcodeDispatcher.cpp
   Interface/Core/SignalDelegator.cpp
   Interface/Core/X86Tables.cpp
@@ -114,6 +118,7 @@ set (SRCS
   Interface/Core/X86Tables/X87Tables.cpp
   Interface/Core/X86Tables/XOPTables.cpp
   Interface/HLE/Thunks/Thunks.cpp
+  Interface/GDBJIT/GDBJIT.cpp
   Interface/IR/AOTIR.cpp
   Interface/IR/IRDumper.cpp
   Interface/IR/IRParser.cpp
@@ -184,7 +189,9 @@ if (ENABLE_JIT_X86_64)
     Interface/Core/JIT/x86_64/MemoryOps.cpp
     Interface/Core/JIT/x86_64/MiscOps.cpp
     Interface/Core/JIT/x86_64/MoveOps.cpp
-    Interface/Core/JIT/x86_64/VectorOps.cpp)
+    Interface/Core/JIT/x86_64/VectorOps.cpp
+    Interface/Core/JIT/x86_64/x64Relocations.cpp
+    )
   list(APPEND DEFINES -DJIT_X86_64)
 endif()
 
@@ -201,7 +208,9 @@ if (ENABLE_JIT_ARM64)
     Interface/Core/JIT/Arm64/MemoryOps.cpp
     Interface/Core/JIT/Arm64/MiscOps.cpp
     Interface/Core/JIT/Arm64/MoveOps.cpp
-    Interface/Core/JIT/Arm64/VectorOps.cpp)
+    Interface/Core/JIT/Arm64/VectorOps.cpp
+    Interface/Core/JIT/Arm64/Arm64Relocations.cpp
+    )
 endif()
 
 set (LIBS vixl dl xxhash tiny-json)
@@ -219,13 +228,11 @@ set(OUTPUT_IR_FOLDER "${CMAKE_BINARY_DIR}/include/FEXCore/IR")
 set(OUTPUT_NAME "${OUTPUT_IR_FOLDER}/IRDefines.inc")
 set(INPUT_NAME "${CMAKE_CURRENT_SOURCE_DIR}/Interface/IR/IR.json")
 
-add_custom_target(CREATE_IR_FOLDER ALL
-  COMMAND ${CMAKE_COMMAND} -E make_directory "${OUTPUT_IR_FOLDER}")
+file(MAKE_DIRECTORY "${OUTPUT_IR_FOLDER}")
 
 add_custom_command(
   OUTPUT "${OUTPUT_NAME}"
   DEPENDS "${INPUT_NAME}"
-  DEPENDS CREATE_IR_FOLDER
   DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py"
   COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py" "${INPUT_NAME}" "${OUTPUT_NAME}"
   )
@@ -239,7 +246,6 @@ set(OUTPUT_IR_DOC "${CMAKE_BINARY_DIR}/IR.md")
 add_custom_command(
   OUTPUT "${OUTPUT_IR_DOC}"
   DEPENDS "${INPUT_NAME}"
-  DEPENDS CREATE_IR_FOLDER
   DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py"
   COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py" "${INPUT_NAME}" "${OUTPUT_IR_DOC}"
   )
@@ -260,15 +266,13 @@ set(INPUT_CONFIG_NAME "${CMAKE_BINARY_DIR}/generated/Config/Config.json")
 set(OUTPUT_MAN_NAME "${CMAKE_BINARY_DIR}/generated/FEX.1")
 set(OUTPUT_MAN_NAME_COMPRESS "${CMAKE_BINARY_DIR}/generated/FEX.1.gz")
 
-add_custom_target(CREATE_CONFIG_FOLDER ALL
-  COMMAND ${CMAKE_COMMAND} -E make_directory "${OUTPUT_CONFIG_FOLDER}")
+file(MAKE_DIRECTORY "${OUTPUT_CONFIG_FOLDER}")
 
 add_custom_command(
   OUTPUT "${OUTPUT_CONFIG_NAME}"
   OUTPUT "${OUTPUT_CONFIG_OPTION_NAME}"
   OUTPUT "${OUTPUT_MAN_NAME}"
   DEPENDS "${INPUT_CONFIG_NAME}"
-  DEPENDS CREATE_CONFIG_FOLDER
   DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/config_generator.py"
   COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/config_generator.py" "${INPUT_CONFIG_NAME}" "${OUTPUT_CONFIG_NAME}" "${OUTPUT_MAN_NAME}"
   "${OUTPUT_CONFIG_OPTION_NAME}"
@@ -329,6 +333,7 @@ function(AddDefaultOptionsToTarget Name)
 
     -Wno-trigraphs
     -ffunction-sections
+    -fwrapv
   )
 
   if (GCC_COLOR)

External/FEXCore/Source/Common/JitSymbols.cpp

@@ -34,6 +34,14 @@ namespace FEXCore {
     fmt::print(fp.get(), "{} {:x} {}_{}\n", HostAddr, CodeSize, Name, HostAddr);
   }
 
+  void JITSymbols::Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name, uintptr_t Offset) {
+    if (!fp) return;
+
+    // Linux perf format is very straightforward
+    // `<HostPtr> <Size> <Name>\n`
+    fmt::print(fp.get(), "{} {:x} {}+0x{:x} ({})\n", HostAddr, CodeSize, Name, Offset, HostAddr);
+  }
+
   void JITSymbols::RegisterNamedRegion(const void *HostAddr, uint32_t CodeSize, std::string_view Name) {
     if (!fp) return;
 

External/FEXCore/Source/Common/JitSymbols.h

@@ -13,6 +13,7 @@ public:
 
   void Register(const void *HostAddr, uint64_t GuestAddr, uint32_t CodeSize);
   void Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name);
+  void Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name, uintptr_t Offset);
   void RegisterNamedRegion(const void *HostAddr, uint32_t CodeSize, std::string_view Name);
   void RegisterJITSpace(const void *HostAddr, uint32_t CodeSize);
 

External/FEXCore/Source/Common/SoftFloat.h

@@ -188,6 +188,10 @@ struct X80SoftFloat {
     return extF80_roundToInt(lhs, softfloat_roundingMode, false);
   }
 
+  static X80SoftFloat FRNDINT(X80SoftFloat const &lhs, uint_fast8_t RoundMode) {
+    return extF80_roundToInt(lhs, RoundMode, false);
+  }
+
   static X80SoftFloat FXTRACT_SIG(X80SoftFloat const &lhs) {
 #if defined(DEBUG_X86_FLOAT)
     BIGFLOAT Result;
@@ -257,7 +261,7 @@ struct X80SoftFloat {
 
     return Result;
 #else
-    X80SoftFloat Int = FRNDINT(rhs);
+    X80SoftFloat Int = FRNDINT(rhs, softfloat_round_minMag);
     BIGFLOAT Src2_d = Int;
     Src2_d = exp2l(Src2_d);
     X80SoftFloat Src2_X80 = Src2_d;

External/FEXCore/Source/Common/StringUtils.h

@@ -0,0 +1,29 @@
+#pragma once
+#include <string>
+
+namespace FEXCore::StringUtils {
+  // Trim the left side of the string of whitespace and new lines
+  [[maybe_unused]] static std::string LeftTrim(std::string String, std::string TrimTokens = " \t\n\r") {
+    size_t pos = std::string::npos;
+    if ((pos = String.find_first_not_of(TrimTokens)) != std::string::npos) {
+      String.erase(0, pos);
+    }
+
+    return String;
+  }
+
+  // Trim the right side of the string of whitespace and new lines
+  [[maybe_unused]] static std::string RightTrim(std::string String, std::string TrimTokens = " \t\n\r") {
+    size_t pos = std::string::npos;
+    if ((pos = String.find_last_not_of(TrimTokens)) != std::string::npos) {
+      String.erase(String.begin() + pos + 1, String.end());
+    }
+
+    return String;
+  }
+
+  // Trim both the left and right of the string of whitespace and new lines
+  [[maybe_unused]] static std::string Trim(std::string String, std::string TrimTokens = " \t\n\r") {
+    return RightTrim(LeftTrim(String, TrimTokens), TrimTokens);
+  }
+}

External/FEXCore/Source/Interface/Config/Config.cpp

@@ -1,4 +1,5 @@
 #include "Common/StringConv.h"
+#include "Common/StringUtils.h"
 #include "Common/Paths.h"
 #include "Utils/FileLoading.h"
 
@@ -370,29 +371,22 @@ namespace JSON {
     return {};
   }
 
-  std::string ltrim(std::string String) {
-    size_t pos = std::string::npos;
-    if ((pos = String.find_first_not_of(" \t\n\r")) != std::string::npos) {
-      String.erase(0, pos);
+
+  std::string FindContainer() {
+    // We only support pressure-vessel at the moment
+    const static std::string ContainerManager = "/run/host/container-manager";
+    if (std::filesystem::exists(ContainerManager)) {
+      std::vector<char> Manager{};
+      if (FEXCore::FileLoading::LoadFile(Manager, ContainerManager)) {
+        // Trim the whitespace, may contain a newline
+        std::string ManagerStr = Manager.data();
+        ManagerStr = FEXCore::StringUtils::Trim(ManagerStr);
+        return ManagerStr;
+      }
     }
-
-    return String;
+    return {};
   }
 
-  std::string rtrim(std::string String) {
-    size_t pos = std::string::npos;
-    if ((pos = String.find_last_not_of(" \t\n\r")) != std::string::npos) {
-      String.erase(String.begin() + pos + 1, String.end());
-    }
-
-    return String;
-  }
-
-  std::string trim(std::string String) {
-    return rtrim(ltrim(String));
-  }
-
-
   std::string FindContainerPrefix() {
     // We only support pressure-vessel at the moment
     const static std::string ContainerManager = "/run/host/container-manager";
@@ -401,7 +395,7 @@ namespace JSON {
       if (FEXCore::FileLoading::LoadFile(Manager, ContainerManager)) {
         // Trim the whitespace, may contain a newline
         std::string ManagerStr = Manager.data();
-        ManagerStr = trim(ManagerStr);
+        ManagerStr = FEXCore::StringUtils::Trim(ManagerStr);
         if (strncmp(ManagerStr.data(), "pressure-vessel", Manager.size()) == 0) {
           // We are running inside of pressure vessel
           // Our $CMAKE_INSTALL_PREFIX paths are now inside of /run/host/$CMAKE_INSTALL_PREFIX
@@ -445,6 +439,16 @@ namespace JSON {
       }
     }
 
+    if (FEXCore::Config::Exists(FEXCore::Config::CONFIG_CACHEOBJECTCODECOMPILATION)) {
+      FEX_CONFIG_OPT(CacheObjectCodeCompilation, CACHEOBJECTCODECOMPILATION);
+      FEX_CONFIG_OPT(Core, CORE);
+
+      if (CacheObjectCodeCompilation() && Core() == FEXCore::Config::CONFIG_INTERPRETER) {
+        // If running the interpreter then disable cache code compilation
+        FEXCore::Config::Erase(FEXCore::Config::CONFIG_CACHEOBJECTCODECOMPILATION);
+      }
+    }
+
     std::string ContainerPrefix { FindContainerPrefix() };
     auto ExpandPathIfExists = [&ContainerPrefix](FEXCore::Config::ConfigOption Config, std::string PathName) {
       auto NewPath = ExpandPath(ContainerPrefix, PathName);

External/FEXCore/Source/Interface/Config/Config.json.in

@@ -38,6 +38,17 @@
           "Number of physical hardware threads to tell the process we have.",
           "0 will auto detect."
         ]
+      },
+      "CacheObjectCodeCompilation": {
+        "Type": "uint32",
+        "Default": "FEXCore::Config::ConfigObjectCodeHandler::CONFIG_NONE",
+        "TextDefault": "none",
+        "Choices": [ "none", "read", "readwrite" ],
+        "ArgumentHandler": "CacheObjectCodeHandler",
+        "Desc": [
+          "Cache JIT object code to drive.",
+          "Allows JIT code to be shared between applications"
+        ]
       }
     },
     "Emulation": {
@@ -104,6 +115,13 @@
           "This can be useful for setting environment variables that thunks can pick up.",
           "Typically isn't necessary since the guest libc isn't thunked. But is possible."
         ]
+      },
+      "AdditionalArguments": {
+        "Type": "strarray",
+        "Default": "",
+        "Desc": [
+          "Allows the user to pass additional arguments to the application"
+        ]
       }
     },
     "Debug": {
@@ -190,6 +208,16 @@
           "Useful for determining hot blocks of code",
           "Has some file writing overhead per JIT block"
         ]
+      },
+      "GDBSymbols": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Integrates with GDB using the JIT interface.",
+          "Needs the fex jit loader in GDB, which can be loaded via `jit-reader-load libFEXGDBReader.so.`",
+          "Also needs x86_64-linux-gnu-objdump in PATH.",
+          "Can be very slow."
+        ]
       }
     },
     "Logging": {
@@ -201,36 +229,28 @@
           "Disables logging"
         ]
       },
-      "OutputSocket": {
-        "Type": "str",
-        "Default": "",
-        "Desc": [
-          "Socket to connect to",
-          "eg: localhost:8087",
-          "If set will override the OutputLog location"
-        ]
-      },
       "OutputLog": {
         "Type": "str",
-        "Default": "stderr",
+        "Default": "server",
         "ShortArg": "o",
         "Desc": [
           "File to write FEX output to.",
-          "[stdout, stderr, <Filename>]"
+          "[stdout, stderr, server, <Filename>]"
         ]
       }
     },
     "Hacks": {
       "SMCChecks": {
         "Type": "uint8",
-        "Default": "FEXCore::Config::CONFIG_SMC_MMAN",
-        "TextDefault": "mman",
+        "Default": "FEXCore::Config::CONFIG_SMC_MTRACK",
+        "TextDefault": "mtrack",
         "ArgumentHandler": "SMCCheckHandler",
         "Desc": [
           "Checks code for modification before execution.",
           "\tnone: No checks",
-          "\tmman: Invalidate on mmap, mprotect, munmap",
-          "\tfull: Validate code before every run (slow)"
+          "\tmtrack: Page tracking based invalidation",
+          "\tfull: Validate code before every run (slow)",
+          "\tmman: Invalidate on mmap, mprotect, munmap (deprecated, use mtrack)"
         ]
       },
       "TSOEnabled": {
@@ -241,6 +261,21 @@
           "Highly likely to break any multithreaded application if disabled."
         ]
       },
+      "TSOAutoMigration": {
+        "Type": "bool",
+        "Default": "true",
+        "Desc": [
+          "Automatically enables TSO when shared memory is used.",
+          "Should work without issues in most cases."
+        ]
+      },
+      "X87ReducedPrecision": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
+        ]
+      },
       "ABILocalFlags": {
         "Type": "bool",
         "Default": "false",

External/FEXCore/Source/Interface/Context/Context.cpp

@@ -43,8 +43,8 @@ namespace FEXCore::Context {
     delete CTX;
   }
 
-  FEXCore::Core::InternalThreadState* InitCore(FEXCore::Context::Context *CTX, FEXCore::CodeLoader *Loader) {
-    return CTX->InitCore(Loader);
+  FEXCore::Core::InternalThreadState* InitCore(FEXCore::Context::Context *CTX, uint64_t InitialRIP, uint64_t StackPointer) {
+    return CTX->InitCore(InitialRIP, StackPointer);
   }
 
   void SetExitHandler(FEXCore::Context::Context *CTX, ExitHandler handler) {
@@ -149,13 +149,14 @@ namespace FEXCore::Context {
   void CleanupAfterFork(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread) {
     CTX->CleanupAfterFork(Thread);
   }
-
+  
   void SetSignalDelegator(FEXCore::Context::Context *CTX, FEXCore::SignalDelegator *SignalDelegation) {
     CTX->SignalDelegation = SignalDelegation;
   }
 
   void SetSyscallHandler(FEXCore::Context::Context *CTX, FEXCore::HLE::SyscallHandler *Handler) {
     CTX->SyscallHandler = Handler;
+    CTX->SourcecodeResolver = Handler->GetSourcecodeResolver();
   }
 
   FEXCore::CPUID::FunctionResults RunCPUIDFunction(FEXCore::Context::Context *CTX, uint32_t Function, uint32_t Leaf) {
@@ -186,11 +187,15 @@ namespace FEXCore::Context {
     CTX->WriteFilesWithCode(Writer);
   }
 
-  void AddNamedRegion(FEXCore::Context::Context *CTX, uintptr_t Base, uintptr_t Length, uintptr_t Offset, const std::string& Name) {
-    return CTX->AddNamedRegion(Base, Length, Offset, Name);
+  IR::AOTIRCacheEntry *LoadAOTIRCacheEntry(FEXCore::Context::Context *CTX, const std::string &Name) {
+    return CTX->LoadAOTIRCacheEntry(Name);
   }
-  void RemoveNamedRegion(FEXCore::Context::Context *CTX, uintptr_t Base, uintptr_t Length) {
-    return CTX->RemoveNamedRegion(Base, Length);
+  void UnloadAOTIRCacheEntry(FEXCore::Context::Context *CTX, IR::AOTIRCacheEntry *Entry) {
+    return CTX->UnloadAOTIRCacheEntry(Entry);
+  }
+
+  CustomIRResult AddCustomIREntrypoint(FEXCore::Context::Context *CTX, uintptr_t Entrypoint, std::function<void(uintptr_t Entrypoint, FEXCore::IR::IREmitter *)> Handler, void *Creator, void *Data) {
+    return CTX->AddCustomIREntrypoint(Entrypoint, Handler, Creator, Data);
   }
 
 namespace Debug {

External/FEXCore/Source/Interface/Context/Context.h

@@ -4,6 +4,8 @@
 #include "Interface/Core/CPUID.h"
 #include "Interface/Core/HostFeatures.h"
 #include "Interface/Core/X86HelperGen.h"
+#include "Interface/Core/ObjectCache/ObjectCacheService.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/IR/AOTIR.h"
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/Context.h>
@@ -34,13 +36,21 @@ class CodeLoader;
 class ThunkHandler;
 class GdbServer;
 
+namespace CodeSerialize {
+  class CodeObjectSerializeService;
+}
+
 namespace CPU {
   class Arm64JITCore;
   class X86JITCore;
+  class InterpreterCore;
+  class Dispatcher;
 }
 namespace HLE {
 struct SyscallArguments;
 class SyscallHandler;
+class SourcecodeResolver;
+struct SourcecodeMap;
 }
 }
 
@@ -67,6 +77,7 @@ namespace FEXCore::Context {
     friend class FEXCore::CPU::X86JITCore;
   #endif
 
+    friend class FEXCore::CPU::InterpreterCore;
     friend class FEXCore::IR::Validation::IRValidation;
 
     struct {
@@ -81,6 +92,7 @@ namespace FEXCore::Context {
       FEX_CONFIG_OPT(GdbServer, GDBSERVER);
       FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
       FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
+      FEX_CONFIG_OPT(TSOAutoMigration, TSOAUTOMIGRATION);
       FEX_CONFIG_OPT(ABILocalFlags, ABILOCALFLAGS);
       FEX_CONFIG_OPT(ABINoPF, ABINOPF);
       FEX_CONFIG_OPT(AOTIRCapture, AOTIRCAPTURE);
@@ -97,16 +109,15 @@ namespace FEXCore::Context {
       FEX_CONFIG_OPT(GlobalJITNaming, GLOBALJITNAMING);
       FEX_CONFIG_OPT(LibraryJITNaming, LIBRARYJITNAMING);
       FEX_CONFIG_OPT(BlockJITNaming, BLOCKJITNAMING);
+      FEX_CONFIG_OPT(GDBSymbols, GDBSYMBOLS);
       FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
+      FEX_CONFIG_OPT(CacheObjectCodeCompilation, CACHEOBJECTCODECOMPILATION);
+      FEX_CONFIG_OPT(x87ReducedPrecision, X87REDUCEDPRECISION);
     } Config;
 
-    using IntCallbackReturn =  FEX_NAKED void(*)(FEXCore::Core::InternalThreadState *Thread, volatile void *Host_RSP);
-    IntCallbackReturn InterpreterCallbackReturn;
-
     FEXCore::HostFeatures HostFeatures;
 
     std::mutex ThreadCreationMutex;
-    uint64_t ThreadID{};
     FEXCore::Core::InternalThreadState* ParentThread;
     std::vector<FEXCore::Core::InternalThreadState*> Threads;
     std::atomic_bool CoreShuttingDown{false};
@@ -118,9 +129,13 @@ namespace FEXCore::Context {
     Event PauseWait;
     bool Running{};
 
+    std::shared_mutex CodeInvalidationMutex; 
+    
     FEXCore::CPUIDEmu CPUID;
     FEXCore::HLE::SyscallHandler *SyscallHandler{};
+    FEXCore::HLE::SourcecodeResolver *SourcecodeResolver{};
     std::unique_ptr<FEXCore::ThunkHandler> ThunkHandler;
+    std::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
 
     CustomCPUFactoryType CustomCPUFactory;
     FEXCore::Context::ExitHandler CustomExitHandler;
@@ -135,7 +150,7 @@ namespace FEXCore::Context {
     Context();
     ~Context();
 
-    FEXCore::Core::InternalThreadState* InitCore(FEXCore::CodeLoader *Loader);
+    FEXCore::Core::InternalThreadState* InitCore(uint64_t InitialRIP, uint64_t StackPointer);
     FEXCore::Context::ExitReason RunUntilExit();
     int GetProgramStatus() const;
     bool IsPaused() const { return !Running; }
@@ -155,13 +170,20 @@ namespace FEXCore::Context {
     void RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
     void RegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
 
-    static void RemoveCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    // Must be called from owning thread
+    static void RemoveThreadCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
 
     // Wrapper which takes CpuStateFrame instead of InternalThreadState
-    static void RemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
-      RemoveCodeEntry(Frame->Thread, GuestRIP);
+    // Must be called from owning thread
+    static void RemoveThreadCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
+      RemoveThreadCodeEntry(Frame->Thread, GuestRIP);
     }
 
+    // returns false if a handler was already registered
+    CustomIRResult AddCustomIREntrypoint(uintptr_t Entrypoint, std::function<void(uintptr_t Entrypoint, FEXCore::IR::IREmitter *)> Handler, void *Creator, void *Data);
+
+    void RemoveCustomIREntrypoint(uintptr_t Entrypoint);
+
     // Debugger interface
     void CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP);
     uint64_t GetThreadCount() const;
@@ -171,21 +193,19 @@ namespace FEXCore::Context {
 
     struct GenerateIRResult {
       FEXCore::IR::IRListView* IRList;
-      // User's responsibility to deallocate this.
-      FEXCore::IR::RegisterAllocationData* RAData;
+      FEXCore::IR::RegisterAllocationData::UniquePtr RAData;
       uint64_t TotalInstructions;
       uint64_t TotalInstructionsLength;
       uint64_t StartAddr;
       uint64_t Length;
     };
-    [[nodiscard]] GenerateIRResult GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    [[nodiscard]] GenerateIRResult GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP, bool ExtendedDebugInfo);
 
     struct CompileCodeResult {
       void* CompiledCode;
       FEXCore::IR::IRListView* IRData;
       FEXCore::Core::DebugData* DebugData;
-      // User's responsibility to deallocate this.
-      FEXCore::IR::RegisterAllocationData* RAData;
+      FEXCore::IR::RegisterAllocationData::UniquePtr RAData;
       bool GeneratedIR;
       uint64_t StartAddr;
       uint64_t Length;
@@ -196,21 +216,9 @@ namespace FEXCore::Context {
     // same as CompileBlock, but aborts on failure
     void CompileBlockJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP);
 
-    /**
-     * @brief Initializes the JIT compilers for the thread
-     *
-     * @param State The internal FEX thread state object
-     * @param CompileThread Is this for the compile service or not?
-     *
-     * InitializeCompiler is called inside of CreateThread, so you likely don't need this
-     * This is exposed because the CompileService needs to initialize compilers while copying data from
-     * the paired InternalThreadState that it is compiling code for
-     */
-    void InitializeCompiler(FEXCore::Core::InternalThreadState* State, bool CompileThread);
-
     // Used for thread creation from syscalls
     /**
-     * @brief Used to create FEX thread objects in preparation for creating a true OS thread
+     * @brief Used to create FEX thread objects in preparation for creating a true OS thread. Does set a TID or PID.
      *
      * @param NewThreadState The initial thread state to setup for our state
      * @param ParentTID The PID that was the parent thread that created this
@@ -233,7 +241,7 @@ namespace FEXCore::Context {
     FEXCore::Core::InternalThreadState* CreateThread(FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID);
 
     /**
-     * @brief Initializes the TLS data for a thread
+     * @brief Initializes TID, PID and TLS data for a thread
      *
      * @param Thread The internal FEX thread state object
      */
@@ -269,8 +277,8 @@ namespace FEXCore::Context {
 
     uint8_t GetGPRSize() const { return Config.Is64BitMode ? 8 : 4; }
 
-    void AddNamedRegion(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename);
-    void RemoveNamedRegion(uintptr_t Base, uintptr_t Size);
+    IR::AOTIRCacheEntry *LoadAOTIRCacheEntry(const std::string &filename);
+    void UnloadAOTIRCacheEntry(IR::AOTIRCacheEntry *Entry);
 
     FEXCore::JITSymbols Symbols;
 
@@ -297,8 +305,15 @@ namespace FEXCore::Context {
       IRCaptureCache.SetAOTIRRenamer(CacheRenamer);
     }
 
+    FEXCore::Utils::PooledAllocatorMMap OpDispatcherAllocator;
+    FEXCore::Utils::PooledAllocatorMMap FrontendAllocator;
+
+    void MarkMemoryShared();
+
+    bool IsTSOEnabled() { return (IsMemoryShared || !Config.TSOAutoMigration) && Config.TSOEnabled; }
+
   protected:
-    void ClearCodeCache(FEXCore::Core::InternalThreadState *Thread, bool AlsoClearIRCache);
+    void ClearCodeCache(FEXCore::Core::InternalThreadState *Thread);
 
   private:
     /**
@@ -310,22 +325,36 @@ namespace FEXCore::Context {
      */
     void InitializeThreadData(FEXCore::Core::InternalThreadState *Thread);
 
+    /**
+     * @brief Initializes the JIT compilers for the thread
+     *
+     * @param State The internal FEX thread state object
+     *
+     * InitializeCompiler is called inside of CreateThread, so you likely don't need this
+     */
+    void InitializeCompiler(FEXCore::Core::InternalThreadState* Thread);
+
     void WaitForIdleWithTimeout();
 
     void NotifyPause();
 
-    void AddBlockMapping(FEXCore::Core::InternalThreadState *Thread, uint64_t Address, void *Ptr, uint64_t Start, uint64_t Length);
-    FEXCore::CodeLoader *LocalLoader{};
+    void AddBlockMapping(FEXCore::Core::InternalThreadState *Thread, uint64_t Address, void *Ptr);
 
     // Entry Cache
-    uint64_t StartingRIP;
     std::mutex ExitMutex;
     std::unique_ptr<GdbServer> DebugServer;
 
     IR::AOTIRCaptureCache IRCaptureCache;
+    std::unique_ptr<FEXCore::CodeSerialize::CodeObjectSerializeService> CodeObjectCacheService;
 
     bool StartPaused = false;
+    bool IsMemoryShared = false;
     FEX_CONFIG_OPT(AppFilename, APP_FILENAME);
+    
+    std::shared_mutex CustomIRMutex;
+    std::unordered_map<uint64_t, std::tuple<std::function<void(uintptr_t Entrypoint, FEXCore::IR::IREmitter *)>, void *, void *>> CustomIRHandlers;
+    FEXCore::CPU::CPUBackendFeatures BackendFeatures;
+    FEXCore::CPU::DispatcherConfig DispatcherConfig;
   };
 
   uint64_t HandleSyscall(FEXCore::HLE::SyscallHandler *Handler, FEXCore::Core::CpuStateFrame *Frame, FEXCore::HLE::SyscallArguments *Args);

External/FEXCore/Source/Interface/Core/ArchHelpers/Arm64.cpp

@@ -1,14 +1,14 @@
 #include "Interface/Core/ArchHelpers/Arm64.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
 
+#include <aarch64/cpu-aarch64.h>
+
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Telemetry.h>
 
 #include <atomic>
-#include <stdint.h>
-
-#include <signal.h>
-#include "aarch64/cpu-aarch64.h"
+#include <csignal>
+#include <cstdint>
 
 namespace FEXCore::ArchHelpers::Arm64 {
 FEXCORE_TELEMETRY_STATIC_INIT(SplitLock, TYPE_HAS_SPLIT_LOCKS);
@@ -513,7 +513,8 @@ uint64_t HandleCASPAL_ARMv8(void *_ucontext, void *_info, uint32_t Instr) {
   //Only 32-bit pairs
   for(int i = 1; i < 10; i++) {
     uint32_t NextInstr = PC[i];
-    if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_INST) {
+    if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_INST ||
+        (NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_SHIFT_INST) {
        ExpectedReg1 = GetRmReg(NextInstr);
     } else if ((NextInstr & FEXCore::ArchHelpers::Arm64::CCMP_MASK) == FEXCore::ArchHelpers::Arm64::CCMP_INST) {
        ExpectedReg2 = GetRmReg(NextInstr);
@@ -1579,7 +1580,7 @@ bool HandleAtomicMemOp(void *_ucontext, void *_info, uint32_t Instr) {
   return false;
 }
 
-bool HandleAtomicLoad(void *_ucontext, void *_info, uint32_t Instr) {
+bool HandleAtomicLoad(void *_ucontext, void *_info, uint32_t Instr, int64_t Offset) {
   mcontext_t* mcontext = &reinterpret_cast<ucontext_t*>(_ucontext)->uc_mcontext;
   siginfo_t* info = reinterpret_cast<siginfo_t*>(_info);
 
@@ -1592,7 +1593,7 @@ bool HandleAtomicLoad(void *_ucontext, void *_info, uint32_t Instr) {
   uint32_t ResultReg = Instr & 0b11111;
   uint32_t AddressReg = (Instr >> 5) & 0b11111;
 
-  uint64_t Addr = mcontext->regs[AddressReg];
+  uint64_t Addr = mcontext->regs[AddressReg] + Offset;
 
   if (Size == 2) {
     auto Res = DoLoad16(Addr);
@@ -1622,7 +1623,7 @@ bool HandleAtomicLoad(void *_ucontext, void *_info, uint32_t Instr) {
   return false;
 }
 
-bool HandleAtomicStore(void *_ucontext, void *_info, uint32_t Instr) {
+bool HandleAtomicStore(void *_ucontext, void *_info, uint32_t Instr, int64_t Offset) {
   mcontext_t* mcontext = &reinterpret_cast<ucontext_t*>(_ucontext)->uc_mcontext;
   siginfo_t* info = reinterpret_cast<siginfo_t*>(_info);
 
@@ -1635,7 +1636,7 @@ bool HandleAtomicStore(void *_ucontext, void *_info, uint32_t Instr) {
   uint32_t DataReg = Instr & 0x1F;
   uint32_t AddressReg = (Instr >> 5) & 0b11111;
 
-  uint64_t Addr = mcontext->regs[AddressReg];
+  uint64_t Addr = mcontext->regs[AddressReg] + Offset;
 
   constexpr bool DoRetry = false;
   if (Size == 2) {
@@ -1745,7 +1746,8 @@ static uint64_t HandleCAS_NoAtomics(void *_ucontext, void *_info)
        #endif
        DesiredReg = GetRdReg(NextInstr);
      }
-     else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_INST) {
+     else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_INST ||
+              (NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_SHIFT_INST) {
        ExpectedReg = GetRmReg(NextInstr);
      }
   }
@@ -1828,11 +1830,13 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
   // Scan forward at most five instructions to find our instructions
   for (size_t i = 1; i < 6; ++i) {
     uint32_t NextInstr = PC[i];
-    if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::ADD_INST) {
+    if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::ADD_INST ||
+        (NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::ADD_SHIFT_INST) {
       AtomicOp = ExclusiveAtomicPairType::TYPE_ADD;
       DataSourceReg = GetRmReg(NextInstr);
     }
-    else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::SUB_INST) {
+    else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::SUB_INST ||
+             (NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::SUB_SHIFT_INST) {
       uint32_t RnReg = GetRnReg(NextInstr);
       if (RnReg == REGISTER_MASK) {
         // Zero reg means neg
@@ -1843,21 +1847,34 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
       }
       DataSourceReg = GetRmReg(NextInstr);
     }
-    else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_INST) {
-    	return HandleCAS_NoAtomics(_ucontext, _info); //ARMv8.0 CAS
+    else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_INST ||
+             (NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::CMP_SHIFT_INST ) {
+      return HandleCAS_NoAtomics(_ucontext, _info); //ARMv8.0 CAS
     }
     else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::AND_INST) {
       AtomicOp = ExclusiveAtomicPairType::TYPE_AND;
       DataSourceReg = GetRmReg(NextInstr);
     }
+    else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::BIC_INST) {
+      AtomicOp = ExclusiveAtomicPairType::TYPE_BIC;
+      DataSourceReg = GetRmReg(NextInstr);
+    }
     else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::OR_INST) {
       AtomicOp = ExclusiveAtomicPairType::TYPE_OR;
       DataSourceReg = GetRmReg(NextInstr);
     }
+    else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::ORN_INST) {
+      AtomicOp = ExclusiveAtomicPairType::TYPE_ORN;
+      DataSourceReg = GetRmReg(NextInstr);
+    }
     else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::EOR_INST) {
       AtomicOp = ExclusiveAtomicPairType::TYPE_EOR;
       DataSourceReg = GetRmReg(NextInstr);
     }
+    else if ((NextInstr & FEXCore::ArchHelpers::Arm64::ALU_OP_MASK) == FEXCore::ArchHelpers::Arm64::EON_INST) {
+      AtomicOp = ExclusiveAtomicPairType::TYPE_EON;
+      DataSourceReg = GetRmReg(NextInstr);
+    }
     else if ((NextInstr & FEXCore::ArchHelpers::Arm64::STLXR_MASK) == FEXCore::ArchHelpers::Arm64::STLXR_INST) {
 #if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
       // Just double check that the memory destination matches
@@ -1888,40 +1905,53 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
   uint32_t Size = 1 << (Instr >> 30);
 
   constexpr bool DoRetry = true;
+
+  auto NOPExpected = []<typename AtomicType>(AtomicType SrcVal, AtomicType) -> AtomicType {
+    return SrcVal;
+  };
+
+  auto ADDDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return SrcVal + Desired;
+  };
+
+  auto SUBDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return SrcVal - Desired;
+  };
+
+  auto ANDDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return SrcVal & Desired;
+  };
+
+  auto BICDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return SrcVal & ~Desired;
+  };
+
+  auto ORDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return SrcVal | Desired;
+  };
+
+  auto ORNDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return SrcVal | ~Desired;
+  };
+
+  auto EORDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return SrcVal ^ Desired;
+  };
+
+  auto EONDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return SrcVal ^ ~Desired;
+  };
+
+  auto NEGDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return -SrcVal;
+  };
+
+  auto SWAPDesired = []<typename AtomicType>(AtomicType SrcVal, AtomicType Desired) -> AtomicType {
+    return Desired;
+  };
+
   if (Size == 2) {
     using AtomicType = uint16_t;
-    auto NOPExpected = [](AtomicType SrcVal, AtomicType) -> AtomicType {
-      return SrcVal;
-    };
-
-    auto ADDDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal + Desired;
-    };
-
-    auto SUBDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal - Desired;
-    };
-
-    auto ANDDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal & Desired;
-    };
-
-    auto ORDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal | Desired;
-    };
-
-    auto EORDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal ^ Desired;
-    };
-
-    auto NEGDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return -SrcVal;
-    };
-
-    auto SWAPDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return Desired;
-    };
-
     CASDesiredFn<AtomicType> DesiredFunction{};
 
     switch (AtomicOp) {
@@ -1937,12 +1967,21 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
       case ExclusiveAtomicPairType::TYPE_AND:
         DesiredFunction = ANDDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_BIC:
+        DesiredFunction = BICDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_OR:
         DesiredFunction = ORDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_ORN:
+        DesiredFunction = ORNDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_EOR:
         DesiredFunction = EORDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_EON:
+        DesiredFunction = EONDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_NEG:
         DesiredFunction = NEGDesired;
         break;
@@ -1966,38 +2005,6 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
   }
   else if (Size == 4) {
     using AtomicType = uint32_t;
-    auto NOPExpected = [](AtomicType SrcVal, AtomicType) -> AtomicType {
-      return SrcVal;
-    };
-
-    auto ADDDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal + Desired;
-    };
-
-    auto SUBDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal - Desired;
-    };
-
-    auto ANDDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal & Desired;
-    };
-
-    auto ORDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal | Desired;
-    };
-
-    auto EORDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal ^ Desired;
-    };
-
-    auto NEGDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return -SrcVal;
-    };
-
-    auto SWAPDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return Desired;
-    };
-
     CASDesiredFn<AtomicType> DesiredFunction{};
 
     switch (AtomicOp) {
@@ -2013,12 +2020,21 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
       case ExclusiveAtomicPairType::TYPE_AND:
         DesiredFunction = ANDDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_BIC:
+        DesiredFunction = BICDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_OR:
         DesiredFunction = ORDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_ORN:
+        DesiredFunction = ORNDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_EOR:
         DesiredFunction = EORDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_EON:
+        DesiredFunction = EONDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_NEG:
         DesiredFunction = NEGDesired;
         break;
@@ -2042,38 +2058,6 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
   }
   else if (Size == 8) {
     using AtomicType = uint64_t;
-    auto NOPExpected = [](AtomicType SrcVal, AtomicType) -> AtomicType {
-      return SrcVal;
-    };
-
-    auto ADDDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal + Desired;
-    };
-
-    auto SUBDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal - Desired;
-    };
-
-    auto ANDDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal & Desired;
-    };
-
-    auto ORDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal | Desired;
-    };
-
-    auto EORDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return SrcVal ^ Desired;
-    };
-
-    auto NEGDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return -SrcVal;
-    };
-
-    auto SWAPDesired = [](AtomicType SrcVal, AtomicType Desired) -> AtomicType {
-      return Desired;
-    };
-
     CASDesiredFn<AtomicType> DesiredFunction{};
 
     switch (AtomicOp) {
@@ -2089,12 +2073,21 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
       case ExclusiveAtomicPairType::TYPE_AND:
         DesiredFunction = ANDDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_BIC:
+        DesiredFunction = BICDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_OR:
         DesiredFunction = ORDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_ORN:
+        DesiredFunction = ORNDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_EOR:
         DesiredFunction = EORDesired;
         break;
+      case ExclusiveAtomicPairType::TYPE_EON:
+        DesiredFunction = EONDesired;
+        break;
       case ExclusiveAtomicPairType::TYPE_NEG:
         DesiredFunction = NEGDesired;
         break;
@@ -2140,7 +2133,7 @@ bool HandleSIGBUS(bool ParanoidTSO, int Signal, void *info, void *ucontext) {
     if ((Instr & 0x3F'FF'FC'00) == 0x08'DF'FC'00 || // LDAR*
         (Instr & 0x3F'FF'FC'00) == 0x38'BF'C0'00) { // LDAPR*
       if (ParanoidTSO) {
-        if (FEXCore::ArchHelpers::Arm64::HandleAtomicLoad(ucontext, info, Instr)) {
+        if (FEXCore::ArchHelpers::Arm64::HandleAtomicLoad(ucontext, info, Instr, 0)) {
           // Skip this instruction now
           ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) + 4);
           return true;
@@ -2164,7 +2157,7 @@ bool HandleSIGBUS(bool ParanoidTSO, int Signal, void *info, void *ucontext) {
     }
     else if ( (Instr & 0x3F'FF'FC'00) == 0x08'9F'FC'00) { // STLR*
       if (ParanoidTSO) {
-        if (FEXCore::ArchHelpers::Arm64::HandleAtomicStore(ucontext, info, Instr)) {
+        if (FEXCore::ArchHelpers::Arm64::HandleAtomicStore(ucontext, info, Instr, 0)) {
           // Skip this instruction now
           ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) + 4);
           return true;
@@ -2186,6 +2179,60 @@ bool HandleSIGBUS(bool ParanoidTSO, int Signal, void *info, void *ucontext) {
         ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) - 4);
       }
     }
+    else if ((Instr & RCPC2_MASK) == LDAPUR_INST) { // LDAPUR*
+      // Extract the 9-bit offset from the instruction
+      int32_t Offset = static_cast<int32_t>(Instr) << 11 >> 23;
+      if (ParanoidTSO) {
+        if (FEXCore::ArchHelpers::Arm64::HandleAtomicLoad(ucontext, info, Instr, Offset)) {
+          // Skip this instruction now
+          ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) + 4);
+          return true;
+        }
+        else {
+          LogMan::Msg::EFmt("Unhandled JIT SIGBUS LDAPUR*: PC: {} Instruction: 0x{:08x}\n", fmt::ptr(PC), PC[0]);
+          return false;
+        }
+      }
+      else {
+        uint32_t LDUR = 0b0011'1000'0100'0000'0000'0000'0000'0000;
+        LDUR |= Size << 30;
+        LDUR |= AddrReg << 5;
+        LDUR |= DataReg;
+        LDUR |= Instr & (0b1'1111'1111 << 9);
+        PC[-1] = DMB;
+        PC[0] = LDUR;
+        PC[1] = DMB;
+        // Back up one instruction and have another go
+        ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) - 4);
+      }
+    }
+    else if ((Instr & RCPC2_MASK) == STLUR_INST) { // STLUR*
+      // Extract the 9-bit offset from the instruction
+      int32_t Offset = static_cast<int32_t>(Instr) << 11 >> 23;
+      if (ParanoidTSO) {
+        if (FEXCore::ArchHelpers::Arm64::HandleAtomicStore(ucontext, info, Instr, Offset)) {
+          // Skip this instruction now
+          ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) + 4);
+          return true;
+        }
+        else {
+          LogMan::Msg::EFmt("Unhandled JIT SIGBUS LDLUR*: PC: {} Instruction: 0x{:08x}\n", fmt::ptr(PC), PC[0]);
+          return false;
+        }
+      }
+      else {
+        uint32_t STUR = 0b0011'1000'0000'0000'0000'0000'0000'0000;
+        STUR |= Size << 30;
+        STUR |= AddrReg << 5;
+        STUR |= DataReg;
+        STUR |= Instr & (0b1'1111'1111 << 9);
+        PC[-1] = DMB;
+        PC[0] = STUR;
+        PC[1] = DMB;
+        // Back up one instruction and have another go
+        ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) - 4);
+      }
+    }
     else if ((Instr & FEXCore::ArchHelpers::Arm64::LDAXP_MASK) == FEXCore::ArchHelpers::Arm64::LDAXP_INST) { // LDAXP
       //Should be compare and swap pair only. LDAXP not used elsewhere
       uint64_t BytesToSkip = FEXCore::ArchHelpers::Arm64::HandleCASPAL_ARMv8(ucontext, info, Instr);

External/FEXCore/Source/Interface/Core/ArchHelpers/Arm64.h

@@ -12,6 +12,10 @@ namespace FEXCore::ArchHelpers::Arm64 {
   constexpr uint32_t ATOMIC_MEM_MASK = 0x3B200C00;
   constexpr uint32_t ATOMIC_MEM_INST = 0x38200000;
 
+  constexpr uint32_t RCPC2_MASK  = 0x3F'E0'0C'00;
+  constexpr uint32_t LDAPUR_INST = 0x19'40'00'00;
+  constexpr uint32_t STLUR_INST  = 0x19'00'00'00;
+
   constexpr uint32_t LDAXP_MASK = 0xBF'FF'80'00;
   constexpr uint32_t LDAXP_INST = 0x88'7F'80'00;
 
@@ -27,13 +31,19 @@ namespace FEXCore::ArchHelpers::Arm64 {
   constexpr uint32_t CBNZ_MASK = 0x7F'00'00'00;
   constexpr uint32_t CBNZ_INST = 0x35'00'00'00;
 
-  constexpr uint32_t ALU_OP_MASK = 0x7F'00'00'00;
-  constexpr uint32_t ADD_INST    = 0x0B'00'00'00;
-  constexpr uint32_t SUB_INST    = 0x4B'00'00'00;
-  constexpr uint32_t CMP_INST    = 0x6B'00'00'00;
-  constexpr uint32_t AND_INST    = 0x0A'00'00'00;
-  constexpr uint32_t OR_INST     = 0x2A'00'00'00;
-  constexpr uint32_t EOR_INST    = 0x4A'00'00'00;
+  constexpr uint32_t ALU_OP_MASK    = 0x7F'20'00'00;
+  constexpr uint32_t ADD_INST       = 0x0B'00'00'00;
+  constexpr uint32_t SUB_INST       = 0x4B'00'00'00;
+  constexpr uint32_t ADD_SHIFT_INST = 0x0B'20'00'00;
+  constexpr uint32_t SUB_SHIFT_INST = 0x4B'20'00'00;
+  constexpr uint32_t CMP_INST       = 0x6B'00'00'00;
+  constexpr uint32_t CMP_SHIFT_INST = 0x6B'20'00'00;
+  constexpr uint32_t AND_INST       = 0x0A'00'00'00;
+  constexpr uint32_t BIC_INST       = 0x0A'20'00'00;
+  constexpr uint32_t OR_INST        = 0x2A'00'00'00;
+  constexpr uint32_t ORN_INST       = 0x2A'20'00'00;
+  constexpr uint32_t EOR_INST       = 0x4A'00'00'00;
+  constexpr uint32_t EON_INST       = 0x4A'20'00'00;
 
   constexpr uint32_t CCMP_MASK   = 0x7F'E0'0C'10;
   constexpr uint32_t CCMP_INST   = 0x7A'40'00'00;
@@ -46,8 +56,11 @@ namespace FEXCore::ArchHelpers::Arm64 {
     TYPE_ADD,
     TYPE_SUB,
     TYPE_AND,
+    TYPE_BIC,
     TYPE_OR,
+    TYPE_ORN,
     TYPE_EOR,
+    TYPE_EON,
     TYPE_NEG, // This is just a sub with zero. Need to know the differences
   };
 
@@ -83,8 +96,8 @@ namespace FEXCore::ArchHelpers::Arm64 {
     return (Instr >> RM_OFFSET) & REGISTER_MASK;
   }
 
-  bool HandleAtomicLoad(void *_ucontext, void *_info, uint32_t Instr);
-  bool HandleAtomicStore(void *_ucontext, void *_info, uint32_t Instr);
+  bool HandleAtomicLoad(void *_ucontext, void *_info, uint32_t Instr, int64_t Offset);
+  bool HandleAtomicStore(void *_ucontext, void *_info, uint32_t Instr, int64_t Offset);
   bool HandleAtomicLoad128(void *_ucontext, void *_info, uint32_t Instr);
   uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info);
   bool HandleCASPAL(void *_ucontext, void *_info, uint32_t Instr);

External/FEXCore/Source/Interface/Core/ArchHelpers/Arm64Emitter.cpp

@@ -1,22 +1,28 @@
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Context/Context.h"
+#include "Interface/HLE/Thunks/Thunks.h"
 
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
 
-#include "aarch64/cpu-aarch64.h"
-#include "cpu-features.h"
-#include "aarch64/instructions-aarch64.h"
-#include "utils-vixl.h"
+#include <aarch64/cpu-aarch64.h>
+#include <aarch64/instructions-aarch64.h>
+#include <cpu-features.h>
+#include <utils-vixl.h>
 
+#include <array>
 #include <tuple>
+#include <utility>
 
 namespace FEXCore::CPU {
 #define STATE x28
 
 // We want vixl to not allocate a default buffer. Jit and dispatcher will manually create one.
-Arm64Emitter::Arm64Emitter(FEXCore::Context::Context *ctx, size_t size) : vixl::aarch64::Assembler(size, vixl::aarch64::PositionDependentCode) {
+Arm64Emitter::Arm64Emitter(FEXCore::Context::Context *ctx, size_t size)
+  : vixl::aarch64::Assembler(size, vixl::aarch64::PositionDependentCode)
+  , EmitterCTX {ctx} {
   CPU.SetUp();
 
   auto Features = vixl::CPUFeatures::InferFromOS();
@@ -42,12 +48,57 @@ void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant,
   }
 
   int NumMoves = 1;
-  movz(Reg, (Constant) & 0xFFFF, 0);
-  for (int i = 1; i < Segments; ++i) {
+  int RequiredMoveSegments{};
+
+  // Count the number of move segments
+  // We only want to use ADRP+ADD if we have more than 1 segment
+  for (size_t i = 0; i < Segments; ++i) {
     uint16_t Part = (Constant >> (i * 16)) & 0xFFFF;
-    if (Part) {
-      movk(Reg, Part, i * 16);
-      ++NumMoves;
+    if (Part != 0) {
+      ++RequiredMoveSegments;
+    }
+  }
+
+  // ADRP+ADD is specifically optimized in hardware
+  // Check if we can use this
+  auto PC = GetCursorAddress<uint64_t>();
+
+  // PC aligned to page
+  uint64_t AlignedPC = PC & ~0xFFFULL;
+
+  // Offset from aligned PC
+  int64_t AlignedOffset = static_cast<int64_t>(Constant) - static_cast<int64_t>(AlignedPC);
+
+  // If the aligned offset is within the 4GB window then we can use ADRP+ADD
+  // and the number of move segments more than 1
+  if (RequiredMoveSegments > 1 && vixl::IsInt32(AlignedOffset)) {
+    // If this is 4k page aligned then we only need ADRP
+    if ((AlignedOffset & 0xFFF) == 0) {
+      adrp(Reg, AlignedOffset >> 12);
+    }
+    else {
+      // If the constant is within 1MB of PC then we can still use ADR to load in a single instruction
+      // 21-bit signed integer here
+      int64_t SmallOffset = static_cast<int64_t>(Constant) - static_cast<int64_t>(PC);
+      if (vixl::IsInt21(SmallOffset)) {
+        adr(Reg, SmallOffset);
+      }
+      else {
+        // Need to use ADRP + ADD
+        adrp(Reg, AlignedOffset >> 12);
+        add(Reg, Reg, Constant & 0xFFF);
+        NumMoves = 2;
+      }
+    }
+  }
+  else {
+    movz(Reg, (Constant) & 0xFFFF, 0);
+    for (int i = 1; i < Segments; ++i) {
+      uint16_t Part = (Constant >> (i * 16)) & 0xFFFF;
+      if (Part) {
+        movk(Reg, Part, i * 16);
+        ++NumMoves;
+      }
     }
   }
 
@@ -260,23 +311,10 @@ void Arm64Emitter::PopDynamicRegsAndLR() {
   add(sp, sp, SPOffset);
 }
 
-void Arm64Emitter::ResetStack() {
-  if (SpillSlots == 0)
-    return;
-
-  if (IsImmAddSub(SpillSlots * 16)) {
-    add(sp, sp, SpillSlots * 16);
-  } else {
-   // Too big to fit in a 12bit immediate
-   LoadConstant(x0, SpillSlots * 16);
-   add(sp, sp, x0);
-  }
-}
-
 void Arm64Emitter::Align16B() {
   uint64_t CurrentOffset = GetCursorAddress<uint64_t>();
   for (uint64_t i = (16 - (CurrentOffset & 0xF)); i != 0; i -= 4) {
-      nop();
+    nop();
   }
 }
 

External/FEXCore/Source/Interface/Core/ArchHelpers/Arm64Emitter.h

@@ -1,15 +1,19 @@
 #pragma once
 
-#include "aarch64/assembler-aarch64.h"
-#include "aarch64/constants-aarch64.h"
-#include "aarch64/cpu-aarch64.h"
-#include "aarch64/operands-aarch64.h"
-#include "platform-vixl.h"
-#include "FEXCore/Config/Config.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
+#include "Interface/Core/ObjectCache/Relocations.h"
+
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/constants-aarch64.h>
+#include <aarch64/cpu-aarch64.h>
+#include <aarch64/operands-aarch64.h>
+#include <platform-vixl.h>
+
+#include <FEXCore/Config/Config.h>
 
 #include <array>
-#include <stddef.h>
-#include <stdint.h>
+#include <cstddef>
+#include <cstdint>
 #include <utility>
 
 namespace FEXCore::CPU {
@@ -60,6 +64,7 @@ class Arm64Emitter : public vixl::aarch64::Assembler {
 protected:
   Arm64Emitter(FEXCore::Context::Context *ctx, size_t size);
 
+  FEXCore::Context::Context *EmitterCTX;
   vixl::aarch64::CPU CPU;
   void LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant, bool NOPPad = false);
   void SpillStaticRegs(bool FPRs = true, uint32_t GPRSpillMask = ~0U, uint32_t FPRSpillMask = ~0U);
@@ -77,11 +82,8 @@ protected:
   void PushCalleeSavedRegisters();
   void PopCalleeSavedRegisters();
 
-  void ResetStack();
   void Align16B();
 
-  uint32_t SpillSlots{};
-
   FEX_CONFIG_OPT(StaticRegisterAllocation, SRA);
 };
 

External/FEXCore/Source/Interface/Core/CPUBackend.cpp

@@ -0,0 +1,87 @@
+#include "Interface/Context/Context.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
+#include <FEXCore/Core/CPUBackend.h>
+
+namespace FEXCore {
+namespace CPU {
+
+CPUBackend::CPUBackend(FEXCore::Core::InternalThreadState *ThreadState, size_t InitialCodeSize, size_t MaxCodeSize)
+    : ThreadState(ThreadState), InitialCodeSize(InitialCodeSize), MaxCodeSize(MaxCodeSize) {}
+
+CPUBackend::~CPUBackend() {
+  for (auto CodeBuffer : CodeBuffers) {
+    FreeCodeBuffer(CodeBuffer);
+  }
+  CodeBuffers.clear();
+}
+
+auto CPUBackend::GetEmptyCodeBuffer() -> CodeBuffer * {
+  if (ThreadState->CurrentFrame->SignalHandlerRefCounter == 0) {
+    if (CodeBuffers.empty()) {
+      auto NewCodeBuffer = AllocateNewCodeBuffer(InitialCodeSize);
+      EmplaceNewCodeBuffer(NewCodeBuffer);
+    } else {
+      if (CodeBuffers.size() > 1) {
+        // If we have more than one code buffer we are tracking then walk them and delete
+        // This is a cleanup step
+        for (size_t i = 1; i < CodeBuffers.size(); i++) {
+          FreeCodeBuffer(CodeBuffers[i]);
+        }
+        CodeBuffers.resize(1);
+      }
+      // Set the current code buffer to the initial
+      CurrentCodeBuffer = &CodeBuffers[0];
+
+      if (CurrentCodeBuffer->Size != MaxCodeSize) {
+        FreeCodeBuffer(*CurrentCodeBuffer);
+
+        // Resize the code buffer and reallocate our code size
+        CurrentCodeBuffer->Size *= 1.5;
+        CurrentCodeBuffer->Size = std::min(CurrentCodeBuffer->Size, MaxCodeSize);
+
+        *CurrentCodeBuffer = AllocateNewCodeBuffer(CurrentCodeBuffer->Size);
+      }
+    }
+  } else {
+    // We have signal handlers that have generated code
+    // This means that we can not safely clear the code at this point in time
+    // Allocate some new code buffers that we can switch over to instead
+    auto NewCodeBuffer = AllocateNewCodeBuffer(InitialCodeSize);
+    EmplaceNewCodeBuffer(NewCodeBuffer);
+  }
+
+  return CurrentCodeBuffer;
+}
+
+auto CPUBackend::AllocateNewCodeBuffer(size_t Size) -> CodeBuffer {
+  CodeBuffer Buffer;
+  Buffer.Size = Size;
+  Buffer.Ptr = static_cast<uint8_t *>(
+      FEXCore::Allocator::mmap(nullptr, Buffer.Size, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  LOGMAN_THROW_A_FMT(!!Buffer.Ptr, "Couldn't allocate code buffer");
+  
+  if (ThreadState->CTX->Config.GlobalJITNaming()) {
+    ThreadState->CTX->Symbols.RegisterJITSpace(Buffer.Ptr, Buffer.Size);
+  }
+  return Buffer;
+}
+
+void CPUBackend::FreeCodeBuffer(CodeBuffer Buffer) {
+  FEXCore::Allocator::munmap(Buffer.Ptr, Buffer.Size);
+}
+
+bool CPUBackend::IsAddressInCodeBuffer(uintptr_t Address) const {
+  for (auto &Buffer: CodeBuffers) {
+    auto start = (uintptr_t)Buffer.Ptr;
+    auto end = start + Buffer.Size;
+
+    if (Address >= start && Address < end) {
+      return true;
+    }
+  }
+
+  return false;
+}
+
+}
+}

External/FEXCore/Source/Interface/Core/CPUID.cpp

@@ -416,7 +416,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
 
   Res.ecx =
     (1 <<  0) | // SSE3
-    (0 <<  1) | // PCLMULQDQ
+    (1 <<  1) | // PCLMULQDQ
     (1 <<  2) | // DS area supports 64bit layout
     (1 <<  3) | // MWait
     (0 <<  4) | // DS-CPL
@@ -446,7 +446,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
     (SUPPORTS_AVX << 28) | // AVX
     (0 << 29) | // F16C
     (CTX->HostFeatures.SupportsRAND << 30) | // RDRAND
-    (0 << 31);  // Hypervisor always returns zero
+    (1 << 31);  // Hypervisor always returns one
 
   Res.edx =
     (1 <<  0) | // FPU
@@ -658,7 +658,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
       (0 << 26) | // Reserved
       (0 << 27) | // Reserved
       (0 << 28) | // Reserved
-      (0 << 29) | // SHA instructions
+      (1 << 29) | // SHA instructions
       (0 << 30) | // Reserved
       (0 << 31);  // Reserved
 
@@ -826,7 +826,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0000h(uint32_t Leaf) {
   // CPUID documentation information:
   // 4000_0000h - 4FFF_FFFFh - No existing or future CPU will return information in this range
   // Reserved entirely for VMs to do whatever they want.
-  Res.eax = 0x40000000;
+  Res.eax = 0x40000001;
 
   // EBX, EDX, ECX become the hypervisor ID signature
   constexpr static char HypervisorID[12] = "FEXIFEXIEMU";
@@ -834,6 +834,25 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0000h(uint32_t Leaf) {
   return Res;
 }
 
+// Hypervisor CPUID information leaf
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0001h(uint32_t Leaf) {
+  FEXCore::CPUID::FunctionResults Res{};
+  if (Leaf == 0) {
+    // EAX[3:0] Is the host architecture that FEX is running under
+#ifdef _M_X86_64
+    // EAX[3:0] = 1 = x86_64 host architecture
+    Res.eax |= 0b0001;
+#elif defined(_M_ARM_64)
+    // EAX[3:0] = 2 = AArch64 host architecture
+    Res.eax |= 0b0010;
+#else
+    // EAX[3:0] = 0 = Unknown architecture
+#endif
+  }
+
+  return Res;
+}
+
 // Highest extended function implemented
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0000h(uint32_t Leaf) {
   FEXCore::CPUID::FunctionResults Res{};
@@ -1228,6 +1247,7 @@ void CPUIDEmu::Init(FEXCore::Context::Context *ctx) {
 #endif
   // Hypervisor CPUID information leaf
   RegisterFunction(0x4000'0000, &CPUIDEmu::Function_4000_0000h);
+  RegisterFunction(0x4000'0001, &CPUIDEmu::Function_4000_0001h);
 
   // Largest extended function number
   RegisterFunction(0x8000'0000, &CPUIDEmu::Function_8000_0000h);

External/FEXCore/Source/Interface/Core/CPUID.h

@@ -80,6 +80,7 @@ private:
   FEXCore::CPUID::FunctionResults Function_15h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_1Ah(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_4000_0000h(uint32_t Leaf);
+  FEXCore::CPUID::FunctionResults Function_4000_0001h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0000h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0001h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0002h(uint32_t Leaf);

External/FEXCore/Source/Interface/Core/CompileService.cpp

@@ -1,165 +0,0 @@
-#include "Interface/Context/Context.h"
-#include "Interface/Core/LookupCache.h"
-#include "Interface/Core/CompileService.h"
-#include "Interface/Core/OpcodeDispatcher.h"
-#include "FEXCore/Debug/InternalThreadState.h"
-#include "FEXCore/HLE/Linux/ThreadManagement.h"
-#include "Interface/IR/PassManager.h"
-
-#include <FEXCore/Core/CPUBackend.h>
-#include <FEXCore/Core/CoreState.h>
-#include <FEXCore/Core/SignalDelegator.h>
-#include <FEXCore/Utils/Event.h>
-#include <FEXCore/Utils/LogManager.h>
-#include <FEXCore/Utils/Threads.h>
-
-#include <memory>
-#include <pthread.h>
-#include <stdio.h>
-
-namespace FEXCore {
-  static void* ThreadHandler(void *Arg) {
-    FEXCore::CompileService *This = reinterpret_cast<FEXCore::CompileService*>(Arg);
-    This->ExecutionThread();
-    return nullptr;
-  }
-
-  CompileService::CompileService(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
-    : CTX {ctx}
-    , ParentThread {Thread} {
-
-    CompileThreadData = std::make_unique<FEXCore::Core::InternalThreadState>();
-    CompileThreadData->IsCompileService = true;
-
-    // We need a compiler for this work thread
-    CTX->InitializeCompiler(CompileThreadData.get(), true);
-    CompileThreadData->CPUBackend->CopyNecessaryDataForCompileThread(ParentThread->CPUBackend.get());
-
-    uint64_t OldMask = FEXCore::Threads::SetSignalMask(~0ULL);
-    WorkerThread = FEXCore::Threads::Thread::Create(ThreadHandler, this);
-    FEXCore::Threads::SetSignalMask(OldMask);
-  }
-
-  void CompileService::Initialize() {
-    // Share CompileService which = this
-    CompileThreadData->CompileService = ParentThread->CompileService;
-  }
-
-  void CompileService::Shutdown() {
-    ShuttingDown = true;
-    // Kick the working thread
-    StartWork.NotifyAll();
-    WorkerThread->join(nullptr);
-  }
-
-  void CompileService::ClearCache(FEXCore::Core::InternalThreadState *Thread) {
-    // On cache clear we need to spin down the execution thread to ensure it isn't trying to give us more work items
-    if (CompileMutex.try_lock()) {
-      // We can only clear these things if we pulled the compile mutex
-
-      // Grab the work queue and clear it
-      // We don't need to grab the queue mutex since this thread will no longer receive any work events
-      // Threads are bounded 1:1
-      while (!WorkQueue.empty()) {
-        WorkQueue.pop();
-      }
-
-      // Go through the garbage collection array and clear it
-      // It's safe to clear things that aren't marked safe since we are clearing cache
-      GCArray.clear();
-
-      LOGMAN_THROW_A_FMT(CompileThreadData->LocalIRCache.empty(), "Compile service must never have LocalIRCache");
-
-      CompileMutex.unlock();
-    }
-
-    // Clear the inverse cache of what is calling us from the Context ClearCache routine
-    auto SelectedThread = Thread->IsCompileService ? ParentThread : Thread;
-    SelectedThread->LookupCache->ClearCache();
-    SelectedThread->CPUBackend->ClearCache();
-  }
-
-  CompileService::WorkItem *CompileService::CompileCode(uint64_t RIP) {
-    WorkItem* ResultItem = nullptr;
-
-    {
-      // Tell the worker thread to compile code for us
-      auto Item = std::make_unique<WorkItem>();
-      Item->RIP = RIP;
-
-      // Fill the threads work queue
-      std::scoped_lock lk(QueueMutex);
-      ResultItem = WorkQueue.emplace(std::move(Item)).get();
-    }
-
-    // Notify the thread that it has more work
-    StartWork.NotifyAll();
-
-    return ResultItem;
-  }
-
-  void CompileService::ExecutionThread() {
-    // Set our thread name so we can see its relation
-    char ThreadName[16]{};
-    snprintf(ThreadName, 16, "%ld-CS", ParentThread->ThreadManager.TID.load());
-    pthread_setname_np(pthread_self(), ThreadName);
-
-    while (true) {
-      // Wait for work
-      StartWork.Wait();
-      if (ShuttingDown.load()) {
-        break;
-      }
-
-      std::scoped_lock lk(CompileMutex);
-      size_t WorkItems{};
-
-      do {
-        // Grab a work item
-        std::unique_ptr<WorkItem> Item{};
-        {
-          std::scoped_lock lk(QueueMutex);
-          WorkItems = WorkQueue.size();
-          if (WorkItems != 0) {
-            Item = std::move(WorkQueue.front());
-            WorkQueue.pop();
-          }
-        }
-
-        // If we had a work item then work on it
-        if (Item) {
-          // Make sure it's not in lookup cache by accident
-          LOGMAN_THROW_A_FMT(CompileThreadData->LookupCache->FindBlock(Item->RIP) == 0, "Compile Service must never have entries in the LookupCache");
-
-          // Code isn't in cache, compile now
-          // Set our thread state's RIP
-          CompileThreadData->CurrentFrame->State.rip = Item->RIP;
-
-          auto [CodePtr, IRList, DebugData, RAData, Generated, StartAddr, Length] = CTX->CompileCode(CompileThreadData.get(), Item->RIP);
-
-          LOGMAN_THROW_A_FMT(Generated == true, "Compile Service doesn't have IR Cache");
-
-          if (!CodePtr) {
-            // XXX: We currently have the expectation that compile service code will be significantly smaller than regular thread's code
-            ERROR_AND_DIE_FMT("Couldn't compile code for thread at RIP: 0x{:x}", Item->RIP);
-          }
-
-          Item->CodePtr = CodePtr;
-          Item->IRList = IRList;
-          Item->DebugData = DebugData;
-          Item->RAData = RAData;
-          Item->StartAddr = StartAddr;
-          Item->Length = Length;
-
-          auto& GCItem = GCArray.emplace_back(std::move(Item));
-          GCItem->ServiceWorkDone.NotifyAll();
-        }
-      } while (WorkItems != 0);
-
-      // Clean up any safe entries in our GC array if we have any.
-      std::erase_if(GCArray, [](const auto& Entry) {
-        return Entry->SafeToClear.load(std::memory_order_relaxed);
-      });
-    }
-  }
-}

External/FEXCore/Source/Interface/Core/CompileService.h

@@ -1,68 +0,0 @@
-#pragma once
-
-#include <FEXCore/Debug/InternalThreadState.h>
-#include <FEXCore/Utils/Event.h>
-#include <FEXCore/Utils/Threads.h>
-
-#include <atomic>
-#include <memory>
-#include <mutex>
-#include <queue>
-#include <stdint.h>
-#include <vector>
-
-namespace FEXCore {
-namespace Context {
-  struct Context;
-}
-namespace IR {
-  class IRListView;
-  class RegisterAllocationData;
-};
-class CompileService final {
-  public:
-    CompileService(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread);
-    void Initialize();
-    void Shutdown();
-
-    struct WorkItem {
-      // Incoming
-      uint64_t RIP{};
-
-      // Outgoing
-      void *CodePtr{};
-      FEXCore::IR::IRListView *IRList{};
-      FEXCore::IR::RegisterAllocationData *RAData{};
-      FEXCore::Core::DebugData *DebugData{};
-      uint64_t StartAddr;
-      uint64_t Length;
-
-      // Communication
-      Event ServiceWorkDone{};
-      std::atomic_bool SafeToClear{};
-    };
-
-    WorkItem *CompileCode(uint64_t RIP);
-    void ClearCache(FEXCore::Core::InternalThreadState *Thread);
-
-    // Public for threading
-    void ExecutionThread();
-
-    bool IsAddressInJITCode(uint64_t Address) const {
-      return CompileThreadData->CPUBackend->IsAddressInJITCode(Address, false, false);
-    }
-  private:
-    FEXCore::Context::Context *CTX;
-    FEXCore::Core::InternalThreadState *ParentThread;
-
-    std::unique_ptr<FEXCore::Threads::Thread> WorkerThread;
-    std::unique_ptr<FEXCore::Core::InternalThreadState> CompileThreadData;
-
-    std::mutex QueueMutex{};
-    std::mutex CompileMutex{};
-    std::queue<std::unique_ptr<WorkItem>> WorkQueue{};
-    std::vector<std::unique_ptr<WorkItem>> GCArray{};
-    Event StartWork{};
-    std::atomic_bool ShuttingDown{false};
-};
-}

External/FEXCore/Source/Interface/Core/Dispatcher/Arm64Dispatcher.cpp

@@ -16,20 +16,23 @@
 #include <array>
 #include <bit>
 #include <cmath>
+#include <cstddef>
 #include <cstdint>
 #include <memory>
-#include <stddef.h>
 
-#include "aarch64/assembler-aarch64.h"
-#include "aarch64/constants-aarch64.h"
-#include "aarch64/operands-aarch64.h"
-#include "aarch64/cpu-aarch64.h"
-#include "code-buffer-vixl.h"
-#include "platform-vixl.h"
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/constants-aarch64.h>
+#include <aarch64/cpu-aarch64.h>
+#include <aarch64/operands-aarch64.h>
+#include <code-buffer-vixl.h>
+#include <platform-vixl.h>
 
 #include <sys/syscall.h>
 #include <unistd.h>
 
+#define STATE_PTR(STATE_TYPE, FIELD) \
+  MemOperand(STATE, offsetof(FEXCore::Core::STATE_TYPE, FIELD))
+
 namespace FEXCore::CPU {
 
 using namespace vixl;
@@ -38,12 +41,12 @@ using namespace vixl::aarch64;
 static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;
 #define STATE x28
 
-Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config)
-  : Dispatcher(ctx, Thread), Arm64Emitter(ctx, MAX_DISPATCHER_CODE_SIZE) {
-  SRAEnabled = config.StaticRegisterAssignment;
+Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config)
+  : FEXCore::CPU::Dispatcher(ctx), Arm64Emitter(ctx, MAX_DISPATCHER_CODE_SIZE)
+  , config(config) {
   SetAllowAssembler(true);
 
-  DispatchPtr = GetCursorAddress<CPUBackend::AsmDispatch>();
+  DispatchPtr = GetCursorAddress<AsmDispatch>();
 
   // while (true) {
   //    Ptr = FindBlock(RIP)
@@ -53,12 +56,9 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   //    Ptr();
   // }
 
-  Literal l_PagePtr {Thread->LookupCache->GetPagePointer()};
   Literal l_CTX {reinterpret_cast<uintptr_t>(CTX)};
   Literal l_Sleep {reinterpret_cast<uint64_t>(SleepThread)};
   Literal l_CompileBlock {GetCompileBlockPtr()};
-  Literal l_ExitFunctionLink {config.ExitFunctionLink};
-  Literal l_ExitFunctionLinkThis {config.ExitFunctionLinkThis};
 
   // Push all the register we need to save
   PushCalleeSavedRegisters();
@@ -71,11 +71,11 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   // Save this stack pointer so we can cleanly shutdown the emulation with a long jump
   // regardless of where we were in the stack
   add(x0, sp, 0);
-  str(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)));
+  str(x0, STATE_PTR(CpuStateFrame, ReturningStackLocation));
 
   AbsoluteLoopTopAddressFillSRA = GetCursorAddress<uint64_t>();
 
-  if (SRAEnabled) {
+  if (config.StaticRegisterAllocation) {
     FillStaticRegs();
   }
 
@@ -92,11 +92,11 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
 
   // Load in our RIP
   // Don't modify x2 since it contains our RIP once the block doesn't exist
-  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
+  ldr(x2, STATE_PTR(CpuStateFrame, State.rip));
   auto RipReg = x2;
 
   // L1 Cache
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.L1Pointer)));
+  ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
 
   and_(x3, RipReg, LookupCache::L1_ENTRIES_MASK);
   add(x0, x0, Operand(x3, Shift::LSL, 4));
@@ -104,21 +104,17 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   cmp(x0, RipReg);
   b(&FullLookup, Condition::ne);
 
-  if (!config.ExecuteBlocksWithCall) {
-    br(x3);
-  } else {
-    b(&CallBlock);
-  }
+  br(x3);
 
   // L1C check failed, do a full lookup
   bind(&FullLookup);
 
   // This is the block cache lookup routine
   // It matches what is going on it LookupCache.h::FindBlock
-  ldr(x0, &l_PagePtr);
+  ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
 
   // Mask the address by the virtual address size so we can check for aliases
-  uint64_t VirtualMemorySize = Thread->LookupCache->GetVirtualMemorySize();
+  uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
   if (std::popcount(VirtualMemorySize) == 1) {
     and_(x3, RipReg, VirtualMemorySize - 1);
   }
@@ -157,44 +153,21 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     // If we've made it here then we have a real compiled block
     {
       // update L1 cache
-      ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.L1Pointer)));
+      ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
 
       and_(x1, RipReg, LookupCache::L1_ENTRIES_MASK);
       add(x0, x0, Operand(x1, Shift::LSL, 4));
       stp(x3, x2, MemOperand(x0));
 
       // Jump to the block
-      if (!config.ExecuteBlocksWithCall) {
-        br(x3);
-      } else {
-        bind(&CallBlock);
-        mov(x0, STATE);
-        blr(x3);
-
-        if (CTX->GetGdbServerStatus()) {
-          // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
-          // This happens when single stepping
-
-          static_assert(sizeof(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-          ldr(x0, &l_CTX);
-          ldr(w0, MemOperand(x0, offsetof(FEXCore::Context::Context, Config.RunningMode)));
-          // If the value == 0 then branch to the top
-          cbz(x0, &LoopTop);
-          // Else we need to pause now
-          b(&ThreadPauseHandler);
-        } else {
-          // Unconditionally loop to the top
-          // We will only stop on error when compiling a block or signal
-          b(&LoopTop);
-        }
-      }
+      br(x3);
     }
   }
 
   {
     bind(&ExitSpillSRA);
     ThreadStopHandlerAddressSpillSRA = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     ThreadStopHandlerAddress = GetCursorAddress<uint64_t>();
@@ -209,7 +182,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   constexpr bool SignalSafeCompile = true;
   {
     ExitFunctionLinkerAddress = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     if (SignalSafeCompile) {
@@ -231,11 +204,10 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
       svc(0);
     }
 
-    ldr(x0, &l_ExitFunctionLinkThis);
-    mov(x1, STATE);
-    mov(x2, lr);
+    mov(x0, STATE);
+    mov(x1, lr);
 
-    ldr(x3, &l_ExitFunctionLink);
+    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
     blr(x3);
 
     if (SignalSafeCompile) {
@@ -256,7 +228,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
       mov(x0, x4);
     }
 
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       FillStaticRegs();
     br(x0);
   }
@@ -265,7 +237,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   {
     bind(&NoBlock);
 
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     if (SignalSafeCompile) {
@@ -312,7 +284,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
       add(sp, sp, 16);
     }
 
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       FillStaticRegs();
 
     b(&LoopTop);
@@ -331,7 +303,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     // Needs to be distinct from the SignalHandlerReturnAddress
     UnimplementedInstructionAddress = GetCursorAddress<uint64_t>();
 
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     hlt(0);
@@ -342,18 +314,17 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     // Needs to be distinct from the SignalHandlerReturnAddress
     OverflowExceptionInstructionAddress = GetCursorAddress<uint64_t>();
 
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
-    LoadConstant(x0, reinterpret_cast<uint64_t>(&SynchronousFaultData));
     LoadConstant(w1, 1);
-    strb(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, FaultToTopAndGeneratedException)));
+    strb(w1, STATE_PTR(CpuStateFrame, SynchronousFaultData.FaultToTopAndGeneratedException));
     LoadConstant(w1, X86State::X86_TRAPNO_OF);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, TrapNo)));
+    str(w1, STATE_PTR(CpuStateFrame, SynchronousFaultData.TrapNo));
     LoadConstant(w1, 0x80);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, si_code)));
+    str(w1, STATE_PTR(CpuStateFrame, SynchronousFaultData.si_code));
     LoadConstant(x1, 0);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, err_code)));
+    str(w1, STATE_PTR(CpuStateFrame, SynchronousFaultData.err_code));
 
     // hlt/udf = SIGILL
     // brk = SIGTRAP
@@ -364,7 +335,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
 
   {
     ThreadPauseHandlerAddressSpillSRA = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     bind(&ThreadPauseHandler);
@@ -399,7 +370,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     // On return to the thunk, the thunk can get whatever its return value is from the thread context depending on ABI handling on its end
     // When the thunk itself returns, it'll do its regular return logic there
     // void ReentrantCallback(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP);
-    CallbackPtr = GetCursorAddress<CPUBackend::JITCallback>();
+    CallbackPtr = GetCursorAddress<JITCallback>();
 
     // We expect the thunk to have previously pushed the registers it was using
     PushCalleeSavedRegisters();
@@ -408,40 +379,112 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     mov(STATE, x0);
 
     // Make sure to adjust the refcounter so we don't clear the cache now
-    LoadConstant(x0, reinterpret_cast<uint64_t>(&SignalHandlerRefCounter));
-    ldr(w2, MemOperand(x0));
+    ldr(w2, STATE_PTR(CpuStateFrame, SignalHandlerRefCounter));
     add(w2, w2, 1);
-    str(w2, MemOperand(x0));
+    str(w2, STATE_PTR(CpuStateFrame, SignalHandlerRefCounter));
 
     // Now push the callback return trampoline to the guest stack
     // Guest will be misaligned because calling a thunk won't correct the guest's stack once we call the callback from the host
     LoadConstant(x0, CTX->X86CodeGen.CallbackReturn);
 
-    ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
+    ldr(x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
     sub(x2, x2, 16);
-    str(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
+    str(x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
 
     // Store the trampoline to the guest stack
     // Guest stack is now correctly misaligned after a regular call instruction
     str(x0, MemOperand(x2));
 
     // Store RIP to the context state
-    str(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
+    str(x1, STATE_PTR(CpuStateFrame, State.rip));
 
     // load static regs
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       FillStaticRegs();
 
     // Now go back to the regular dispatcher loop
     b(&LoopTop);
   }
 
-  place(&l_PagePtr);
+  {
+    LUDIVHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR();
+
+    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUDIV));
+
+    SpillStaticRegs();
+    blr(x3);
+    FillStaticRegs();
+
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LDIVHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR();
+
+    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LDIV));
+
+    SpillStaticRegs();
+    blr(x3);
+    FillStaticRegs();
+
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LUREMHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR();
+
+    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUREM));
+
+    SpillStaticRegs();
+    blr(x3);
+    FillStaticRegs();
+
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LREMHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR();
+
+    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LREM));
+
+    SpillStaticRegs();
+    blr(x3);
+    FillStaticRegs();
+
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
   place(&l_CTX);
   place(&l_Sleep);
   place(&l_CompileBlock);
-  place(&l_ExitFunctionLink);
-  place(&l_ExitFunctionLinkThis);
 
 
   FinalizeCode();
@@ -457,51 +500,114 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   if (CTX->Config.GlobalJITNaming()) {
     CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));
   }
-
-  // Setup dispatcher specific pointers that need to be accessed from JIT code
-  {
-    auto &Pointers = ThreadState->CurrentFrame->Pointers.AArch64;
-
-    Pointers.DispatcherLoopTop = AbsoluteLoopTopAddress;
-    Pointers.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
-    Pointers.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddressSpillSRA;
-    Pointers.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddressSpillSRA;
-    Pointers.UnimplementedInstructionHandler = UnimplementedInstructionAddress;
-    Pointers.OverflowExceptionHandler = OverflowExceptionInstructionAddress;
-    Pointers.SignalReturnHandler = SignalHandlerReturnAddress;
-    Pointers.L1Pointer = Thread->LookupCache->GetL1Pointer();
-  }
 }
 
-void Arm64Dispatcher::SpillSRA(void *ucontext, uint32_t IgnoreMask) {
+// Used by GenerateGDBPauseCheck, GenerateInterpreterTrampoline, destination buffer is set before use
+static thread_local vixl::aarch64::Assembler emit((uint8_t*)&emit, 1);
+
+size_t Arm64Dispatcher::GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) {
+
+  *emit.GetBuffer() = vixl::CodeBuffer(CodeBuffer, MaxGDBPauseCheckSize);
+  
+  vixl::CodeBufferCheckScope scope(&emit, MaxGDBPauseCheckSize, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
+
+  aarch64::Label RunBlock;
+
+  // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
+  // This happens when single stepping
+
+  static_assert(sizeof(FEXCore::Context::Context::Config.RunningMode) == 4, "This is expected to be size of 4");
+  emit.ldr(x0, STATE_PTR(CpuStateFrame, Thread)); // Get thread
+  emit.ldr(x0, MemOperand(x0, offsetof(FEXCore::Core::InternalThreadState, CTX))); // Get Context
+  emit.ldr(w0, MemOperand(x0, offsetof(FEXCore::Context::Context, Config.RunningMode)));
+
+  // If the value == 0 then we don't need to stop
+  emit.cbz(w0, &RunBlock);
+  {
+    Literal l_GuestRIP {GuestRIP};
+    // Make sure RIP is syncronized to the context
+    emit.ldr(x0, &l_GuestRIP);
+    emit.str(x0, STATE_PTR(CpuStateFrame, State.rip));
+
+    // Stop the thread
+    emit.ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA));
+    emit.br(x0);
+    emit.place(&l_GuestRIP);
+  }
+  emit.bind(&RunBlock);
+  emit.FinalizeCode();
+
+  auto UsedBytes = emit.GetBuffer()->GetCursorOffset();
+  vixl::aarch64::CPU::EnsureIAndDCacheCoherency(CodeBuffer, UsedBytes);
+  return UsedBytes;
+}
+
+size_t Arm64Dispatcher::GenerateInterpreterTrampoline(uint8_t *CodeBuffer) {
+  LOGMAN_THROW_A_FMT(!config.StaticRegisterAllocation, "GenerateInterpreterTrampoline dispatcher does not support SRA");
+  
+  *emit.GetBuffer() = vixl::CodeBuffer(CodeBuffer, MaxInterpreterTrampolineSize);
+
+  vixl::CodeBufferCheckScope scope(&emit, MaxInterpreterTrampolineSize, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
+
+  aarch64::Label InlineIRData;
+
+  emit.mov(x0, STATE);
+  emit.adr(x1, &InlineIRData);
+
+  emit.ldr(x3, STATE_PTR(CpuStateFrame, Pointers.Interpreter.FragmentExecuter));
+  emit.blr(x3);
+
+  emit.ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+  emit.br(x0);
+
+  emit.bind(&InlineIRData);
+
+  emit.FinalizeCode();
+
+  auto UsedBytes = emit.GetBuffer()->GetCursorOffset();
+  vixl::aarch64::CPU::EnsureIAndDCacheCoherency(CodeBuffer, UsedBytes);
+  return UsedBytes;
+}
+
+void Arm64Dispatcher::SpillSRA(FEXCore::Core::InternalThreadState *Thread, void *ucontext, uint32_t IgnoreMask) {
   for(int i = 0; i < SRA64.size(); i++) {
     if (IgnoreMask & (1U << SRA64[i].GetCode())) {
       // Skip this one, it's already spilled
       continue;
     }
-    ThreadState->CurrentFrame->State.gregs[i] = ArchHelpers::Context::GetArmReg(ucontext, SRA64[i].GetCode());
+    Thread->CurrentFrame->State.gregs[i] = ArchHelpers::Context::GetArmReg(ucontext, SRA64[i].GetCode());
   }
 
   for(int i = 0; i < SRAFPR.size(); i++) {
     auto FPR = ArchHelpers::Context::GetArmFPR(ucontext, SRAFPR[i].GetCode());
-    memcpy(&ThreadState->CurrentFrame->State.xmm[i][0], &FPR, sizeof(__uint128_t));
+    memcpy(&Thread->CurrentFrame->State.xmm[i][0], &FPR, sizeof(__uint128_t));
   }
 }
 
-#ifdef _M_ARM_64
+void Arm64Dispatcher::InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) {
+// Setup dispatcher specific pointers that need to be accessed from JIT code
+  {
+    auto &Common = Thread->CurrentFrame->Pointers.Common;
 
-void InterpreterCore::CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
-  DispatcherConfig config;
-  config.ExecuteBlocksWithCall = true;
+    Common.DispatcherLoopTop = AbsoluteLoopTopAddress;
+    Common.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
+    Common.ExitFunctionLinker = ExitFunctionLinkerAddress;
+    Common.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddressSpillSRA;
+    Common.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddressSpillSRA;
+    Common.UnimplementedInstructionHandler = UnimplementedInstructionAddress;
+    Common.OverflowExceptionHandler = OverflowExceptionInstructionAddress;
+    Common.SignalReturnHandler = SignalHandlerReturnAddress;
 
-  Dispatcher = std::make_unique<Arm64Dispatcher>(ctx, Thread, config);
-  DispatchPtr = Dispatcher->DispatchPtr;
-  CallbackPtr = Dispatcher->CallbackPtr;
-
-  // TODO: It feels wrong to initialize this way
-  ctx->InterpreterCallbackReturn = Dispatcher->ReturnPtr;
+    auto &AArch64 = Thread->CurrentFrame->Pointers.AArch64;
+    AArch64.LUDIVHandler = LUDIVHandlerAddress;
+    AArch64.LDIVHandler = LDIVHandlerAddress;
+    AArch64.LUREMHandler = LUREMHandlerAddress;
+    AArch64.LREMHandler = LREMHandlerAddress;
+  }
 }
 
-#endif
-
+std::unique_ptr<Dispatcher> Dispatcher::CreateArm64(FEXCore::Context::Context *CTX, const DispatcherConfig &Config) {
+  return std::make_unique<Arm64Dispatcher>(CTX, Config);
 }
+
+}

External/FEXCore/Source/Interface/Core/Dispatcher/Arm64Dispatcher.h

@@ -15,10 +15,21 @@ namespace FEXCore::CPU {
 
 class Arm64Dispatcher final : public Dispatcher, public Arm64Emitter {
   public:
-    Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config);
+    Arm64Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config);
+    void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) override;
+    size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) override;
+    size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) override;
 
   protected:
-    void SpillSRA(void *ucontext, uint32_t IgnoreMask) override;
+    void SpillSRA(FEXCore::Core::InternalThreadState *Thread, void *ucontext, uint32_t IgnoreMask) override;
+
+  private:
+    // Long division helpers
+    uint64_t LUDIVHandlerAddress{};
+    uint64_t LDIVHandlerAddress{};
+    uint64_t LUREMHandlerAddress{};
+    uint64_t LREMHandlerAddress{};
+    DispatcherConfig config;
 };
 
 }

External/FEXCore/Source/Interface/Core/Dispatcher/Dispatcher.cpp

@@ -1,8 +1,8 @@
 
+#include "Interface/Context/Context.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/X86HelperGen.h"
-#include "Interface/Core/CompileService.h"
 
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CoreState.h>
@@ -40,7 +40,7 @@ void Dispatcher::SleepThread(FEXCore::Context::Context *ctx, FEXCore::Core::CpuS
   ctx->IdleWaitCV.notify_all();
 }
 
-ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(int Signal, void *ucontext) {
+ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(FEXCore::Core::InternalThreadState *Thread, int Signal, void *ucontext) {
   // We can end up getting a signal at any point in our host state
   // Jump to a handler that saves all state so we can safely return
   uint64_t OldSP = ArchHelpers::Context::GetSp(ucontext);
@@ -65,7 +65,7 @@ ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(int Signal, vo
   // Save guest state
   // We can't guarantee if registers are in context or host GPRs
   // So we need to save everything
-  memcpy(&Context->GuestState, ThreadState->CurrentFrame, sizeof(FEXCore::Core::CPUState));
+  memcpy(&Context->GuestState, Thread->CurrentFrame, sizeof(FEXCore::Core::CPUState));
 
   // Set the new SP
   ArchHelpers::Context::SetSp(ucontext, NewSP);
@@ -82,13 +82,13 @@ ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(int Signal, vo
   Context->SigInfoLocation = 0;
 
   // Store fault to top status and then reset it
-  Context->FaultToTopAndGeneratedException = SynchronousFaultData.FaultToTopAndGeneratedException;
-  SynchronousFaultData.FaultToTopAndGeneratedException = false;
+  Context->FaultToTopAndGeneratedException = Thread->CurrentFrame->SynchronousFaultData.FaultToTopAndGeneratedException;
+  Thread->CurrentFrame->SynchronousFaultData.FaultToTopAndGeneratedException = false;
 
   return Context;
 }
 
-void Dispatcher::RestoreThreadState(void *ucontext) {
+void Dispatcher::RestoreThreadState(FEXCore::Core::InternalThreadState *Thread, void *ucontext) {
   uint64_t OldSP{};
   if (CTX->Config.Core() == FEXCore::Config::CONFIG_IRJIT) {
     OldSP = ArchHelpers::Context::GetSp(ucontext);
@@ -103,13 +103,13 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
   auto Context = reinterpret_cast<ArchHelpers::Context::ContextBackup*>(NewSP);
 
   // First thing, reset the guest state
-  memcpy(ThreadState->CurrentFrame, &Context->GuestState, sizeof(FEXCore::Core::CPUState));
+  memcpy(Thread->CurrentFrame, &Context->GuestState, sizeof(FEXCore::Core::CPUState));
 
   // Now restore host state
   ArchHelpers::Context::RestoreContext(ucontext, Context);
 
   if (Context->UContextLocation) {
-    auto Frame = ThreadState->CurrentFrame;
+    auto Frame = Thread->CurrentFrame;
 
     if (Context->Flags &ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_INJIT) {
       // XXX: Unsupported since it needs state reconstruction
@@ -133,7 +133,7 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
         Frame->State.rip = guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_RIP];
         // XXX: Full context setting
         uint32_t eflags = guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_EFL];
-        for (size_t i = 0; i < 32; ++i) {
+        for (size_t i = 0; i < Core::CPUState::NUM_EFLAG_BITS; ++i) {
           Frame->State.flags[i] = (eflags & (1U << i)) ? 1 : 0;
         }
 
@@ -191,7 +191,7 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
         // XXX: Full context setting
         // First 32-bytes of flags is EFLAGS broken out
         uint32_t eflags = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EFL];
-        for (size_t i = 0; i < 32; ++i) {
+        for (size_t i = 0; i < Core::CPUState::NUM_EFLAG_BITS; ++i) {
           Frame->State.flags[i] = (eflags & (1U << i)) ? 1 : 0;
         }
 
@@ -219,7 +219,7 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
         FEXCore::x86::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86::_libc_fpstate*>(guest_uctx->uc_mcontext.fpregs);
 
         // Copy float registers
-        for (size_t i = 0; i < 8; ++i) {
+        for (size_t i = 0; i < Core::CPUState::NUM_MMS; ++i) {
           // 32-bit st register size is only 10 bytes. Not padded to 16byte like x86-64
           memcpy(&Frame->State.mm[i], &fpstate->_st[i], 10);
         }
@@ -274,14 +274,14 @@ static uint32_t ConvertSignalToError(int Signal, siginfo_t *HostSigInfo) {
   return 0;
 }
 
-bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) {
-  auto ContextBackup = StoreThreadState(Signal, ucontext);
+bool Dispatcher::HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) {
+  auto ContextBackup = StoreThreadState(Thread, Signal, ucontext);
 
-  auto Frame = ThreadState->CurrentFrame;
+  auto Frame = Thread->CurrentFrame;
 
   // Ref count our faults
   // We use this to track if it is safe to clear cache
-  ++SignalHandlerRefCounter;
+  ++Thread->CurrentFrame->SignalHandlerRefCounter;
 
   uint64_t OldPC = ArchHelpers::Context::GetPc(ucontext);
   // Set the new PC
@@ -300,7 +300,7 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
   // We are going to be returning to the top of the dispatcher which will fill again
   // Otherwise we might load garbage
   if (SRAEnabled) {
-    if (IsAddressInJITCode(OldPC, false)) {
+    if (Thread->CPUBackend->IsAddressInCodeBuffer(OldPC)) {
       uint32_t IgnoreMask{};
 #ifdef _M_ARM_64
       if (Frame->InSyscallInfo != 0) {
@@ -324,11 +324,11 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
 #endif
 
       // We are in jit, SRA must be spilled
-      SpillSRA(ucontext, IgnoreMask);
+      SpillSRA(Thread, ucontext, IgnoreMask);
 
       ContextBackup->Flags |= ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_INJIT;
     } else {
-      if (!IsAddressInJITCode(OldPC, true)) {
+      if (!IsAddressInDispatcher(OldPC)) {
         // This is likely to cause issues but in some cases it isn't fatal
         // This can also happen if we have put a signal on hold, then we just reenabled the signal
         // So we are in the syscall handler
@@ -410,11 +410,11 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       *guest_siginfo = *HostSigInfo;
 
       if (ContextBackup->FaultToTopAndGeneratedException) {
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = SynchronousFaultData.TrapNo;
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = SynchronousFaultData.err_code;
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = Frame->SynchronousFaultData.TrapNo;
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = Frame->SynchronousFaultData.err_code;
 
         // Overwrite si_code
-        guest_siginfo->si_code = SynchronousFaultData.si_code;
+        guest_siginfo->si_code = Thread->CurrentFrame->SynchronousFaultData.si_code;
       }
       else {
         guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = ConvertSignalToTrapNo(Signal, HostSigInfo);
@@ -501,9 +501,9 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES] = Frame->State.es;
       guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS] = Frame->State.ds;
       if (ContextBackup->FaultToTopAndGeneratedException) {
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = SynchronousFaultData.TrapNo;
-        guest_siginfo->si_code = SynchronousFaultData.si_code;
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = SynchronousFaultData.err_code;
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = Frame->SynchronousFaultData.TrapNo;
+        guest_siginfo->si_code = Frame->SynchronousFaultData.si_code;
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = Frame->SynchronousFaultData.err_code;
       }
       else {
         guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = ConvertSignalToTrapNo(Signal, HostSigInfo);
@@ -529,7 +529,7 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
 #undef COPY_REG
 
       // Copy float registers
-      for (size_t i = 0; i < 8; ++i) {
+      for (size_t i = 0; i < Core::CPUState::NUM_MMS; ++i) {
         // 32-bit st register size is only 10 bytes. Not padded to 16byte like x86-64
         memcpy(&fpstate->_st[i], &Frame->State.mm[i], 10);
       }
@@ -634,44 +634,44 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
   return true;
 }
 
-bool Dispatcher::HandleSIGILL(int Signal, void *info, void *ucontext) {
+bool Dispatcher::HandleSIGILL(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) {
 
   if (ArchHelpers::Context::GetPc(ucontext) == SignalHandlerReturnAddress) {
-    RestoreThreadState(ucontext);
+    RestoreThreadState(Thread, ucontext);
 
     // Ref count our faults
     // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
+    --Thread->CurrentFrame->SignalHandlerRefCounter;
     return true;
   }
 
   if (ArchHelpers::Context::GetPc(ucontext) == PauseReturnInstruction) {
-    RestoreThreadState(ucontext);
+    RestoreThreadState(Thread, ucontext);
 
     // Ref count our faults
     // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
+    --Thread->CurrentFrame->SignalHandlerRefCounter;
     return true;
   }
 
   return false;
 }
 
-bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
-  FEXCore::Core::SignalEvent SignalReason = ThreadState->SignalReason.load();
-  auto Frame = ThreadState->CurrentFrame;
+bool Dispatcher::HandleSignalPause(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) {
+  FEXCore::Core::SignalEvent SignalReason = Thread->SignalReason.load();
+  auto Frame = Thread->CurrentFrame;
 
   if (SignalReason == FEXCore::Core::SignalEvent::Pause) {
     // Store our thread state so we can come back to this
-    StoreThreadState(Signal, ucontext);
+    StoreThreadState(Thread, Signal, ucontext);
 
-    if (SRAEnabled && IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), false)) {
+    if (SRAEnabled && Thread->CPUBackend->IsAddressInCodeBuffer(ArchHelpers::Context::GetPc(ucontext))) {
       // We are in jit, SRA must be spilled
       ArchHelpers::Context::SetPc(ucontext, ThreadPauseHandlerAddressSpillSRA);
     } else {
       if (SRAEnabled) {
         // We are in non-jit, SRA is already spilled
-        LOGMAN_THROW_A_FMT(!IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), true),
+        LOGMAN_THROW_A_FMT(!IsAddressInDispatcher(ArchHelpers::Context::GetPc(ucontext)),
                            "Signals in dispatcher have unsynchronized context");
       }
       ArchHelpers::Context::SetPc(ucontext, ThreadPauseHandlerAddress);
@@ -682,9 +682,9 @@ bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
 
     // Ref count our faults
     // We use this to track if it is safe to clear cache
-    ++SignalHandlerRefCounter;
+    ++Thread->CurrentFrame->SignalHandlerRefCounter;
 
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
+    Thread->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
     return true;
   }
 
@@ -695,16 +695,16 @@ bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
     ArchHelpers::Context::SetSp(ucontext, Frame->ReturningStackLocation);
 
     // Our ref counting doesn't matter anymore
-    SignalHandlerRefCounter = 0;
+    Thread->CurrentFrame->SignalHandlerRefCounter = 0;
 
     // Set the new PC
-    if (SRAEnabled && IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), false)) {
+    if (SRAEnabled && Thread->CPUBackend->IsAddressInCodeBuffer(ArchHelpers::Context::GetPc(ucontext))) {
       // We are in jit, SRA must be spilled
       ArchHelpers::Context::SetPc(ucontext, ThreadStopHandlerAddressSpillSRA);
     } else {
       if (SRAEnabled) {
         // We are in non-jit, SRA is already spilled
-        LOGMAN_THROW_A_FMT(!IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), true),
+        LOGMAN_THROW_A_FMT(!IsAddressInDispatcher(ArchHelpers::Context::GetPc(ucontext)),
                            "Signals in dispatcher have unsynchronized context");
       }
       ArchHelpers::Context::SetPc(ucontext, ThreadStopHandlerAddress);
@@ -713,24 +713,24 @@ bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
     // We need to be a little bit careful here
     // If we were already paused (due to GDB) and we are immediately stopping (due to gdb kill)
     // Then we need to ensure we don't double decrement our idle thread counter
-    if (ThreadState->RunningEvents.ThreadSleeping) {
+    if (Thread->RunningEvents.ThreadSleeping) {
       // If the thread was sleeping then its idle counter was decremented
       // Reincrement it here to not break logic
-      ++ThreadState->CTX->IdleWaitRefCount;
+      ++Thread->CTX->IdleWaitRefCount;
     }
 
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
+    Thread->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
     return true;
   }
 
   if (SignalReason == FEXCore::Core::SignalEvent::Return) {
-    RestoreThreadState(ucontext);
+    RestoreThreadState(Thread, ucontext);
 
     // Ref count our faults
     // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
+    --Thread->CurrentFrame->SignalHandlerRefCounter;
 
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
+    Thread->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
     return true;
   }
 
@@ -749,31 +749,4 @@ uint64_t Dispatcher::GetCompileBlockPtr() {
   return CompileBlockPtr.Data;
 }
 
-void Dispatcher::RemoveCodeBuffer(uint8_t* start_to_remove) {
-  for (auto iter = CodeBuffers.begin(); iter != CodeBuffers.end(); ++iter) {
-    auto [start, end] = *iter;
-    if (start == reinterpret_cast<uint64_t>(start_to_remove)) {
-      CodeBuffers.erase(iter);
-      return;
-    }
-  }
-}
-
-bool Dispatcher::IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher, bool IncludeCompileService) const {
-  for (auto [start, end] : CodeBuffers) {
-    if (Address >= start && Address < end) {
-      return true;
-    }
-  }
-
-  if (IncludeDispatcher && IsAddressInDispatcher(Address)) {
-    return true;
-  }
-
-  if (IncludeCompileService &&  ThreadState->CompileService && ThreadState->CompileService->IsAddressInJITCode(Address)) {
-    return true;
-  }
-  return false;
-}
-
 }

External/FEXCore/Source/Interface/Core/Dispatcher/Dispatcher.h

@@ -1,8 +1,6 @@
 #pragma once
 
 #include <FEXCore/Core/CPUBackend.h>
-
-#include "Interface/Context/Context.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
 
 #include <cstdint>
@@ -21,22 +19,20 @@ struct CpuStateFrame;
 struct InternalThreadState;
 }
 
+namespace FEXCore::Context {
+struct Context;
+}
+
 namespace FEXCore::CPU {
 
 struct DispatcherConfig {
-  bool ExecuteBlocksWithCall = false;
-  uintptr_t ExitFunctionLink = 0;
-  uintptr_t ExitFunctionLinkThis = 0;
-  bool StaticRegisterAssignment = false;
+  bool StaticRegisterAllocation = false;
 };
 
 class Dispatcher {
 public:
   virtual ~Dispatcher() = default;
-  CPUBackend::AsmDispatch DispatchPtr;
-  CPUBackend::JITCallback CallbackPtr;
-  FEXCore::Context::Context::IntCallbackReturn ReturnPtr;
-
+  
   /**
    * @name Dispatch Helper functions
    * @{ */
@@ -50,59 +46,66 @@ public:
   uint64_t SignalHandlerReturnAddress{};
   uint64_t UnimplementedInstructionAddress{};
   uint64_t OverflowExceptionInstructionAddress{};
+  uint64_t IntCallbackReturnAddress{};
 
   uint64_t PauseReturnInstruction{};
 
   /**  @} */
 
-  uint32_t SignalHandlerRefCounter{};
-  struct SynchronousFaultDataStruct {
-    bool FaultToTopAndGeneratedException{};
-    uint32_t TrapNo;
-    uint32_t err_code;
-    uint32_t si_code;
-  } SynchronousFaultData;
-
   uint64_t Start{};
   uint64_t End{};
 
-  bool HandleGuestSignal(int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack);
-  bool HandleSIGILL(int Signal, void *info, void *ucontext);
-  bool HandleSignalPause(int Signal, void *info, void *ucontext);
+  bool HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack);
+  bool HandleSIGILL(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext);
+  bool HandleSignalPause(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext);
 
-  void RegisterCodeBuffer(uint8_t* start, size_t size) {
-    CodeBuffers.emplace_back(reinterpret_cast<uint64_t>(start),
-        reinterpret_cast<uint64_t>(start + size));
-  }
-
-  void RemoveCodeBuffer(uint8_t* start);
-
-  bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true, bool IncludeCompileService = true) const;
   bool IsAddressInDispatcher(uint64_t Address) const {
     return Address >= Start && Address < End;
   }
 
-protected:
-  Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
-    : CTX {ctx}
-    , ThreadState {Thread} {}
+  virtual void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) = 0;
 
-  ArchHelpers::Context::ContextBackup* StoreThreadState(int Signal, void *ucontext);
-  void RestoreThreadState(void *ucontext);
+  // These are across all arches for now
+  static constexpr size_t MaxGDBPauseCheckSize = 128;
+  static constexpr size_t MaxInterpreterTrampolineSize = 128;
+
+  virtual size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) = 0;
+  virtual size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) = 0;
+
+  static std::unique_ptr<Dispatcher> CreateX86(FEXCore::Context::Context *CTX, const DispatcherConfig &Config);
+  static std::unique_ptr<Dispatcher> CreateArm64(FEXCore::Context::Context *CTX, const DispatcherConfig &Config);
+  
+  void ExecuteDispatch(FEXCore::Core::CpuStateFrame *Frame) {
+    DispatchPtr(Frame);
+  }
+
+  void ExecuteJITCallback(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP) {
+    CallbackPtr(Frame, RIP);
+  }
+
+protected:
+  Dispatcher(FEXCore::Context::Context *ctx)
+    : CTX {ctx}
+    {}
+
+  ArchHelpers::Context::ContextBackup* StoreThreadState(FEXCore::Core::InternalThreadState *Thread, int Signal, void *ucontext);
+  void RestoreThreadState(FEXCore::Core::InternalThreadState *Thread, void *ucontext);
   std::stack<uint64_t, std::vector<uint64_t>> SignalFrames;
 
   bool SRAEnabled = false;
-  virtual void SpillSRA(void *ucontext, uint32_t IgnoreMask) {}
+  virtual void SpillSRA(FEXCore::Core::InternalThreadState *Thread, void *ucontext, uint32_t IgnoreMask) {}
 
   FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *ThreadState;
 
   static void SleepThread(FEXCore::Context::Context *ctx, FEXCore::Core::CpuStateFrame *Frame);
 
   static uint64_t GetCompileBlockPtr();
 
-private:
-  std::vector<std::tuple<uint64_t, uint64_t>> CodeBuffers; // Start, End
+  using AsmDispatch = void(*)(FEXCore::Core::CpuStateFrame *Frame);
+  using JITCallback = void(*)(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP);
+
+  AsmDispatch DispatchPtr;
+  JITCallback CallbackPtr;
 };
 
 }

External/FEXCore/Source/Interface/Core/Dispatcher/X86Dispatcher.cpp

@@ -18,21 +18,26 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <sys/mman.h>
-#include "xbyak/xbyak.h"
+#include <xbyak/xbyak.h>
+
+#define STATE_PTR(STATE_TYPE, FIELD) \
+  [STATE + offsetof(FEXCore::Core::STATE_TYPE, FIELD)]
 
 namespace FEXCore::CPU {
 static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;
 #define STATE r14
 
-X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config)
-  : Dispatcher(ctx, Thread)
+X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config)
+  : Dispatcher(ctx)
   , Xbyak::CodeGenerator(MAX_DISPATCHER_CODE_SIZE,
       FEXCore::Allocator::mmap(nullptr, MAX_DISPATCHER_CODE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0),
       nullptr) {
 
+  LOGMAN_THROW_A_FMT(!config.StaticRegisterAllocation, "X86 dispatcher does not support SRA");
+
   using namespace Xbyak;
   using namespace Xbyak::util;
-  DispatchPtr = getCurr<CPUBackend::AsmDispatch>();
+  DispatchPtr = getCurr<AsmDispatch>();
 
   // Temp registers
   // rax, rcx, rdx, rsi, r8, r9,
@@ -78,11 +83,10 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
 
   // Save this stack pointer so we can cleanly shutdown the emulation with a long jump
   // regardless of where we were in the stack
-  mov(qword [rdi + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)], rsp);
+  mov(qword STATE_PTR(CpuStateFrame, ReturningStackLocation), rsp);
 
   Label LoopTop;
   Label FullLookup;
-  Label CallBlock;
   Label NoBlock;
   Label ExitBlock;
   Label ThreadPauseHandler;
@@ -92,29 +96,24 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
 
   {
     // Load our RIP
-    mov(rdx, qword [STATE + offsetof(FEXCore::Core::CPUState, rip)]);
+    mov(rdx, qword STATE_PTR(CPUState, rip));
 
     // L1 Cache
-    mov(r13, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
     mov(rax, rdx);
 
     and_(rax, LookupCache::L1_ENTRIES_MASK);
     shl(rax, 4);
-    cmp(qword[r13 + rax + 8], rdx);
+    cmp(qword[r13 + rax + offsetof(FEXCore::LookupCache::LookupCacheEntry, GuestCode)], rdx);
     jne(FullLookup);
 
-    if (!config.ExecuteBlocksWithCall) {
-      jmp(qword[r13 + rax + 0]);
-    } else {
-      mov(rax, qword[r13 + rax + 0]);
-      jmp(CallBlock);
-    }
+    jmp(qword[r13 + rax + offsetof(FEXCore::LookupCache::LookupCacheEntry, HostCode)]);
 
     L(FullLookup);
-    mov(r13, Thread->LookupCache->GetPagePointer());
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
 
     // Full lookup
-    uint64_t VirtualMemorySize = Thread->LookupCache->GetVirtualMemorySize();
+    uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
     mov(rax, rdx);
     mov(rbx, VirtualMemorySize - 1);
     and_(rax, rbx);
@@ -143,7 +142,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     je(NoBlock);
 
     // Update L1
-    mov(r13, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
     mov(rcx, rdx);
     and_(rcx, LookupCache::L1_ENTRIES_MASK);
     shl(rcx, 1);
@@ -151,30 +150,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     mov(qword[r13 + rcx*8 + 0], rax);
 
     // Real block if we made it here
-    if (!config.ExecuteBlocksWithCall) {
-      jmp(rax);
-    } else {
-      L(CallBlock);
-      mov(rdi, STATE);
-      call(rax);
-
-      if (CTX->GetGdbServerStatus()) {
-        // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
-        // This happens when single stepping
-        static_assert(sizeof(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-        mov(rax, qword [STATE + (offsetof(FEXCore::Core::InternalThreadState, CTX))]);
-
-        // If the value == 0 then branch to the top
-        cmp(dword [rax + (offsetof(FEXCore::Context::Context, Config.RunningMode))], 0);
-        je(LoopTop);
-        // Else we need to pause now
-        jmp(ThreadPauseHandler);
-        ud2();
-      }
-      else {
-        jmp(LoopTop);
-      }
-    }
+    jmp(rax);
   }
 
   {
@@ -193,10 +169,37 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     ret();
   }
 
+  constexpr bool SignalSafeCompile = true;
   // Block creation
   {
     L(NoBlock);
 
+    if (SignalSafeCompile) {
+      // When compiling code, mask all signals to reduce the chance of reentrant allocations
+      // RDI: SETMASK
+      // RSI: Pointer to mask value (uint64_t)
+      // RDX: Pointer to old mask value (uint64_t)
+      // R10: Size of mask, sizeof(uint64_t)
+      // RAX: Syscall
+
+      // Backup rdx
+      mov(r9, rdx);
+
+      mov(rdi, ~0ULL);
+      sub(rsp, 16);
+      mov(qword [rsp], rdi);
+      mov(qword [rsp + 8], rdi);
+
+      mov(rdi, SIG_SETMASK);
+      mov(rsi, rsp);
+      mov(rdx, rsp);
+      mov(r10, 8);
+      mov(rax, SYS_rt_sigprocmask);
+      syscall();
+
+      mov(rdx, r9);
+    }
+
     // {rdi, rsi, rdx}
     mov(rdi, reinterpret_cast<uint64_t>(CTX));
     mov(rsi, STATE);
@@ -204,20 +207,84 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
 
     call(rax);
 
+    if (SignalSafeCompile) {
+      // Now restore the signal mask
+      // Living in the same location
+      // Backup rdx
+      mov(r9, rdx);
+
+      mov(rdi, SIG_SETMASK);
+      mov(rsi, rsp);
+      mov(rdx, 0); // Don't care about result
+      mov(r10, 8);
+      mov(rax, SYS_rt_sigprocmask);
+      syscall();
+
+      // Bring stack back
+      add(rsp, 16);
+
+      mov(rdx, r9);
+    }
+
     // rdx already contains RIP here
     jmp(LoopTop);
   }
 
   {
     ExitFunctionLinkerAddress = getCurr<uint64_t>();
-    // {rdi, rsi, rdx}
-    mov(rdi, config.ExitFunctionLinkThis);
-    mov(rsi, STATE);
-    mov(rdx, rax); // rax is set at the block end
+    if (SignalSafeCompile) {
+      // When compiling code, mask all signals to reduce the chance of reentrant allocations
+      // RDI: SETMASK
+      // RSI: Pointer to mask value (uint64_t)
+      // RDX: Pointer to old mask value (uint64_t)
+      // R10: Size of mask, sizeof(uint64_t)
+      // RAX: Syscall
 
-    mov(rax, config.ExitFunctionLink);
-    call(rax);
-    jmp(rax);
+      // Backup rax
+      mov(r9, rax);
+
+      mov(rdi, ~0ULL);
+      sub(rsp, 16);
+      mov(qword [rsp], rdi);
+      mov(qword [rsp + 8], rdi);
+
+      mov(rdi, SIG_SETMASK);
+      mov(rsi, rsp);
+      mov(rdx, rsp);
+      mov(r10, 8);
+      mov(rax, SYS_rt_sigprocmask);
+      syscall();
+
+      mov(rax, r9);
+    }
+
+    // {rdi, rsi}
+    mov(rdi, STATE);
+    mov(rsi, rax); // rax is set at the block end
+
+    call(qword STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
+
+    if (SignalSafeCompile) {
+      // Now restore the signal mask
+      // Living in the same location
+      // Backup rax
+      mov(r9, rax);
+
+      mov(rdi, SIG_SETMASK);
+      mov(rsi, rsp);
+      mov(rdx, 0); // Don't care about result
+      mov(r10, 8);
+      mov(rax, SYS_rt_sigprocmask);
+      syscall();
+
+      // Bring stack back
+      add(rsp, 16);
+
+      jmp(r9);
+    }
+    else {
+      jmp(rax);
+    }
   }
 
   {
@@ -237,7 +304,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
   }
 
   {
-    CallbackPtr = getCurr<CPUBackend::JITCallback>();
+    CallbackPtr = getCurr<JITCallback>();
 
     push(rbx);
     push(rbp);
@@ -252,7 +319,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     // XXX: XMM?
 
     // Make sure to adjust the refcounter so we don't clear the cache now
-    add(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalHandlerRefCountPointer)], 1);
+    add(qword STATE_PTR(CpuStateFrame, SignalHandlerRefCounter), 1);
 
     // Now push the callback return trampoline to the guest stack
     // Guest will be misaligned because calling a thunk won't correct the guest's stack once we call the callback from the host
@@ -260,12 +327,12 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
 
     // Store the trampoline to the guest stack
     // Guest stack is now correctly misaligned after a regular call instruction
-    sub(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])], 16);
-    mov(rbx, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])]);
+    sub(qword STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]), 16);
+    mov(rbx, qword STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
     mov(qword [rbx], rax);
 
     // Store RIP to the context state
-    mov(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.rip)], rsi);
+    mov(qword STATE_PTR(CpuStateFrame, State.rip), rsi);
 
     // Back to the loop top now
     jmp(LoopTop);
@@ -292,18 +359,17 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     // ud2 = SIGILL
     // int3 = SIGTRAP
     // hlt = SIGSEGV
-
-    mov(rax, reinterpret_cast<uint64_t>(&SynchronousFaultData));
-    add(byte [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, FaultToTopAndGeneratedException)], 1);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, TrapNo)], X86State::X86_TRAPNO_OF);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, err_code)], 0);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, si_code)], 0x80);
+  
+    add(byte STATE_PTR(CpuStateFrame, SynchronousFaultData.FaultToTopAndGeneratedException), 1);
+    mov(dword STATE_PTR(CpuStateFrame, SynchronousFaultData.TrapNo), X86State::X86_TRAPNO_OF);
+    mov(dword STATE_PTR(CpuStateFrame, SynchronousFaultData.err_code), 0);
+    mov(dword STATE_PTR(CpuStateFrame, SynchronousFaultData.si_code), 0x80);
 
     hlt();
   }
 
   {
-    ReturnPtr = getCurr<FEXCore::Context::Context::IntCallbackReturn>();
+    IntCallbackReturnAddress = getCurr<uint64_t>();
 //  using CallbackReturn =  FEX_NAKED void(*)(FEXCore::Core::InternalThreadState *Thread, volatile void *Host_RSP);
 
     // rdi = thread
@@ -337,39 +403,91 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(Start), End-Start);
   }
 
-  // Setup dispatcher specific pointers that need to be accessed from JIT code
-  {
-    auto &Pointers = ThreadState->CurrentFrame->Pointers.X86;
+}
 
-    Pointers.DispatcherLoopTop = AbsoluteLoopTopAddress;
-    Pointers.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
-    Pointers.ThreadStopHandler = ThreadStopHandlerAddress;
-    Pointers.ThreadPauseHandler = ThreadPauseHandlerAddress;
-    Pointers.UnimplementedInstructionHandler = UnimplementedInstructionAddress;
-    Pointers.OverflowExceptionHandler = OverflowExceptionInstructionAddress;
-    Pointers.SignalReturnHandler = SignalHandlerReturnAddress;
-    Pointers.L1Pointer = Thread->LookupCache->GetL1Pointer();
+// Used by GenerateGDBPauseCheck, GenerateInterpreterTrampoline
+static thread_local Xbyak::CodeGenerator emit(1, &emit); // actual emit target set with setNewBuffer
+
+size_t X86Dispatcher::GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) {
+  using namespace Xbyak;
+  using namespace Xbyak::util;
+
+  emit.setNewBuffer(CodeBuffer, MaxGDBPauseCheckSize);
+  
+  Label RunBlock;
+
+  // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
+  // This happens when single stepping
+  static_assert(sizeof(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
+  emit.mov(rax, reinterpret_cast<uint64_t>(CTX));
+
+  // If the value == 0 then we don't need to stop
+  emit.cmp(dword [rax + (offsetof(FEXCore::Context::Context, Config.RunningMode))], 0);
+  emit.je(RunBlock);
+  {
+    // Make sure RIP is syncronized to the context
+    emit.mov(rax, GuestRIP);
+    emit.mov(qword STATE_PTR(CpuStateFrame, State.rip), rax);
+
+    // Stop the thread
+    emit.mov(rax, qword STATE_PTR(CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA));
+    emit.jmp(rax);
   }
+
+  emit.L(RunBlock);
+
+  emit.ready();
+
+  return emit.getSize();
+}
+
+
+size_t X86Dispatcher::GenerateInterpreterTrampoline(uint8_t *CodeBuffer) {
+  using namespace Xbyak;
+  using namespace Xbyak::util;
+
+  emit.setNewBuffer(CodeBuffer, MaxInterpreterTrampolineSize);
+
+  Label InlineIRData;
+  
+  emit.mov(rdi, STATE);
+  emit.lea(rsi, ptr[rip + InlineIRData]);
+  emit.call(qword STATE_PTR(CpuStateFrame, Pointers.Interpreter.FragmentExecuter));
+
+  emit.jmp(qword STATE_PTR(CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+
+  emit.L(InlineIRData);
+
+  emit.ready();
+
+  return emit.getSize();
 }
 
 X86Dispatcher::~X86Dispatcher() {
   FEXCore::Allocator::munmap(top_, MAX_DISPATCHER_CODE_SIZE);
 }
 
-#ifdef _M_X86_64
+void X86Dispatcher::InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) {
+  // Setup dispatcher specific pointers that need to be accessed from JIT code
+  {
+    auto &Common = Thread->CurrentFrame->Pointers.Common;
 
-void InterpreterCore::CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
-  DispatcherConfig config;
-  config.ExecuteBlocksWithCall = true;
+    Common.DispatcherLoopTop = AbsoluteLoopTopAddress;
+    Common.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
+    Common.ExitFunctionLinker = ExitFunctionLinkerAddress;
+    Common.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddress;
+    Common.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddress;
+    Common.UnimplementedInstructionHandler = UnimplementedInstructionAddress;
+    Common.OverflowExceptionHandler = OverflowExceptionInstructionAddress;
+    Common.SignalReturnHandler = SignalHandlerReturnAddress;
 
-  Dispatcher = std::make_unique<X86Dispatcher>(ctx, Thread, config);
-  DispatchPtr = Dispatcher->DispatchPtr;
-  CallbackPtr = Dispatcher->CallbackPtr;
-
-  // TODO: It feels wrong to initialize this way
-  ctx->InterpreterCallbackReturn = Dispatcher->ReturnPtr;
+    auto &Interpreter = Thread->CurrentFrame->Pointers.Interpreter;
+    (uintptr_t&)Interpreter.CallbackReturn = IntCallbackReturnAddress;
+  }
 }
 
-#endif
+std::unique_ptr<Dispatcher> Dispatcher::CreateX86(FEXCore::Context::Context *CTX, const DispatcherConfig &Config) {
+  return std::make_unique<X86Dispatcher>(CTX, Config);
+}
 
 }

External/FEXCore/Source/Interface/Core/Dispatcher/X86Dispatcher.h

@@ -17,7 +17,10 @@ namespace FEXCore::CPU {
 
 class X86Dispatcher final : public Dispatcher, public Xbyak::CodeGenerator {
   public:
-    X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config);
+    X86Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config);
+    void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) override;
+    size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) override;
+    size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) override;
 
     virtual ~X86Dispatcher() override;
 };

External/FEXCore/Source/Interface/Core/Frontend.cpp

@@ -19,6 +19,7 @@ $end_info$
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Telemetry.h>
+#include <FEXHeaderUtils/TypeDefines.h>
 #include <set>
 #include <sys/mman.h>
 
@@ -177,17 +178,12 @@ static uint32_t MapVEXToReg(uint8_t vvvv, bool HasXMM) {
 
 Decoder::Decoder(FEXCore::Context::Context *ctx)
   : CTX {ctx}
-  , OSABI { ctx->SyscallHandler ? ctx->SyscallHandler->GetOSABI() : FEXCore::HLE::SyscallOSABI::OS_UNKNOWN } {
-  // Using mmap is a start-up time optimization
-  // Take advantage of page faulting to reduce startup time for minimal runtime cost
-  DecodedBuffer =
-    reinterpret_cast<FEXCore::X86Tables::DecodedInst *>(
-      FEXCore::Allocator::mmap(0, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize,
-      PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  , OSABI { ctx->SyscallHandler ? ctx->SyscallHandler->GetOSABI() : FEXCore::HLE::SyscallOSABI::OS_UNKNOWN }
+  , PoolObject {ctx->FrontendAllocator, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize} {
 }
 
 Decoder::~Decoder() {
-  FEXCore::Allocator::munmap(DecodedBuffer, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize);
+  PoolObject.UnclaimBuffer();
 }
 
 uint8_t Decoder::ReadByte() {
@@ -1140,7 +1136,7 @@ const uint8_t *Decoder::AdjustAddrForSpecialRegion(uint8_t const* _InstStream, u
   return _InstStream - EntryPoint + RIP;
 }
 
-void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC) {
+void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC, std::function<void(uint64_t BlockEntry, uint64_t Start, uint64_t Length)> AddContainedCodePage) {
   Blocks.clear();
   BlocksToDecode.clear();
   HasBlocks.clear();
@@ -1148,6 +1144,7 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
   DecodedSize = 0;
   MaxCondBranchForward = 0;
   MaxCondBranchBackwards = ~0ULL;
+  DecodedBuffer = PoolObject.ReownOrClaimBuffer();
 
   // XXX: Load symbol data
   SymbolAvailable = false;
@@ -1169,6 +1166,13 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
   // Entry is a jump target
   BlocksToDecode.emplace(PC);
 
+  uint64_t CurrentCodePage = PC & FHU::FEX_PAGE_MASK;
+
+  std::set<uint64_t> CodePages = { CurrentCodePage };
+
+  AddContainedCodePage(PC, CurrentCodePage, FHU::FEX_PAGE_SIZE);
+  
+
   while (!BlocksToDecode.empty()) {
     auto BlockDecodeIt = BlocksToDecode.begin();
     uint64_t RIPToDecode = *BlockDecodeIt;
@@ -1185,10 +1189,33 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
     InstStream = AdjustAddrForSpecialRegion(_InstStream, EntryPoint, RIPToDecode);
 
     while (1) {
+
+      // MAX_INST_SIZE assumes worst case
+      auto OpMinAddress = RIPToDecode + PCOffset;
+      auto OpMaxAddress = OpMinAddress + MAX_INST_SIZE;
+
+      auto OpMinPage = OpMinAddress & FHU::FEX_PAGE_MASK;
+      auto OpMaxPage = OpMaxAddress & FHU::FEX_PAGE_MASK;
+
+
+      if (OpMinPage != CurrentCodePage) {
+        CurrentCodePage = OpMinPage;
+        if (CodePages.insert(CurrentCodePage).second) {
+          AddContainedCodePage(PC, CurrentCodePage, FHU::FEX_PAGE_SIZE);
+        }
+      }
+
+      if (OpMaxPage != CurrentCodePage) {
+        CurrentCodePage = OpMaxPage;
+        if (CodePages.insert(CurrentCodePage).second) {
+          AddContainedCodePage(PC, CurrentCodePage, FHU::FEX_PAGE_SIZE);
+        }
+      }
+
       bool ErrorDuringDecoding = !DecodeInstruction(RIPToDecode + PCOffset);
 
       if (ErrorDuringDecoding) {
-        LogMan::Msg::DFmt("Couldn't Decode something at 0x{:x}, Started at 0x{:x}", PC + PCOffset, PC);
+        LogMan::Msg::DFmt("Couldn't Decode something at 0x{:x}, Started at 0x{:x}", RIPToDecode + PCOffset, PC);
         // Put an invalid instruction in the stream so the core can raise SIGILL if hit
         CurrentBlockDecoding.HasInvalidInstruction = true;
         // Error while decoding instruction. We don't know the table or instruction size

External/FEXCore/Source/Interface/Core/Frontend.h

@@ -27,7 +27,7 @@ public:
 
   Decoder(FEXCore::Context::Context *ctx);
   ~Decoder();
-  void DecodeInstructionsAtEntry(uint8_t const* InstStream, uint64_t PC);
+  void DecodeInstructionsAtEntry(uint8_t const* InstStream, uint64_t PC, std::function<void(uint64_t BlockEntry, uint64_t Start, uint64_t Length)> AddContainedCodePage);
 
   std::vector<DecodedBlocks> const *GetDecodedBlocks() const {
     return &Blocks;
@@ -38,6 +38,11 @@ public:
 
   void SetSectionMaxAddress(uint64_t v) { SectionMaxAddress = v; }
   void SetExternalBranches(std::set<uint64_t> *v) { ExternalBranches = v; }
+
+  void DelayedDisownBuffer() {
+    PoolObject.DelayedDisownBuffer();
+  }
+
 private:
   // To pass any information from instruction prefixes
   // down into the actual instruction handling machinery.
@@ -63,6 +68,7 @@ private:
 
   static constexpr size_t DefaultDecodedBufferSize = 0x10000;
   FEXCore::X86Tables::DecodedInst *DecodedBuffer{};
+  Utils::FixedSizePooledAllocation<FEXCore::X86Tables::DecodedInst*, 5000, 500> PoolObject;
   size_t DecodedSize {};
 
   uint8_t const *InstStream;

External/FEXCore/Source/Interface/Core/GdbServer.h

@@ -6,8 +6,10 @@ $end_info$
 #pragma once
 
 #include <FEXCore/Config/Config.h>
+#include <FEXCore/Utils/Event.h>
 #include <FEXCore/Utils/Threads.h>
 
+#include <atomic>
 #include <istream>
 #include <memory>
 #include <mutex>
@@ -27,6 +29,10 @@ public:
     // Public for threading
     void GdbServerLoop();
 
+    void AlertLibrariesChanged() {
+      LibraryMapChanged = true;
+    }
+
 private:
     void Break(int signal);
 
@@ -38,6 +44,9 @@ private:
 
     void SendACK(std::ostream &stream, bool NACK);
 
+    Event ThreadBreakEvent{};
+    void WaitForThreadWakeup();
+
     struct HandledPacketType {
       std::string Response{};
       enum ResponseType {
@@ -74,9 +83,13 @@ private:
     bool NoAckMode{false};
     bool NonStopMode{false};
     std::string ThreadString{};
-    std::string MemoryMapString{};
     std::string OSDataString{};
+    void buildLibraryMap();
+    std::atomic<bool> LibraryMapChanged = true;
+    std::string LibraryMapString{};
 
+    // Used to keep track of which signals to pass to the guest
+    std::array<bool, SignalDelegator::MAX_SIGNALS + 1> PassSignals{};
     uint32_t CurrentDebuggingThread{};
     int ListenSocket{};
     FEX_CONFIG_OPT(Filename, APP_FILENAME);

External/FEXCore/Source/Interface/Core/HostFeatures.cpp

@@ -59,11 +59,8 @@ HostFeatures::HostFeatures() {
 
   // Only supported when FEAT_AFP is supported
   SupportsFlushInputsToZero = Features.Has(vixl::CPUFeatures::Feature::kAFP);
-
-  // RCPC is bugged on Snapdragon 865
-  // Causes glibc cond16 test to immediately throw assert
-  // __pthread_mutex_cond_lock: Assertion `mutex->__data.__owner == 0'
-  SupportsRCPC = false; //Features.Has(vixl::CPUFeatures::Feature::kRCpc);
+  SupportsRCPC = Features.Has(vixl::CPUFeatures::Feature::kRCpc);
+  SupportsTSOImm9 = Features.Has(vixl::CPUFeatures::Feature::kRCpcImm);
 
   // We need to get the CPU's cache line size
   // We expect sane targets that have correct cacheline sizes across clusters
@@ -83,6 +80,18 @@ HostFeatures::HostFeatures() {
   SupportsAES = Features.has(Xbyak::util::Cpu::tAESNI);
   SupportsCRC = Features.has(Xbyak::util::Cpu::tSSE42);
   SupportsRAND = Features.has(Xbyak::util::Cpu::tRDRAND) && Features.has(Xbyak::util::Cpu::tRDSEED);
+  SupportsRCPC = true;
+  SupportsTSOImm9 = true;
+
+  // xbyak doesn't know how to check for CLZero
+  uint32_t eax, ebx, ecx, edx;
+  // First ensure we support a new enough extended CPUID function range
+  __cpuid(0x8000'0000, eax, ebx, ecx, edx);
+  if (eax >= 0x8000'0008U) {
+    // CLZero defined in 8000_00008_EBX[bit 0]
+    __cpuid(0x8000'0008, eax, ebx, ecx, edx);
+    SupportsCLZERO = ebx & 1;
+  }
 
   SupportsFlushInputsToZero = true;
   SupportsFloatExceptions = true;

External/FEXCore/Source/Interface/Core/HostFeatures.h

@@ -19,6 +19,7 @@ class HostFeatures final {
     bool SupportsCLZERO{};
     bool SupportsAtomics{};
     bool SupportsRCPC{};
+    bool SupportsTSOImm9{};
     bool SupportsRAND{};
 
     // Float exception behaviour

External/FEXCore/Source/Interface/Core/Interpreter/BranchOps.cpp

@@ -4,6 +4,7 @@ tags: backend|interpreter
 $end_info$
 */
 
+#include "Interface/Context/Context.h"
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
@@ -25,20 +26,13 @@ static void SignalReturn(FEXCore::Core::InternalThreadState *Thread) {
 }
 
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
 
 DEF_OP(SignalReturn) {
   SignalReturn(Data->State);
 }
 
 DEF_OP(CallbackReturn) {
-  Data->State->CTX->InterpreterCallbackReturn(Data->State, Data->StackEntry);
+  Data->State->CurrentFrame->Pointers.Interpreter.CallbackReturn(Data->State, Data->StackEntry);
 }
 
 DEF_OP(ExitFunction) {
@@ -147,8 +141,8 @@ DEF_OP(ValidateCode) {
   }
 }
 
-DEF_OP(RemoveCodeEntry) {
-  Data->State->CTX->RemoveCodeEntry(Data->State, Data->CurrentEntry);
+DEF_OP(RemoveThreadCodeEntry) {
+  Data->State->CTX->RemoveThreadCodeEntry(Data->State, Data->CurrentEntry);
 }
 
 DEF_OP(CPUID) {

External/FEXCore/Source/Interface/Core/Interpreter/EncryptionOps.cpp

@@ -513,6 +513,44 @@ DEF_OP(CRC32) {
   memcpy(GDP, &Tmp, sizeof(Tmp));
 }
 
+DEF_OP(PCLMUL) {
+  auto Op = IROp->C<IR::IROp_PCLMUL>();
+
+  const auto Selector = Op->Selector;
+  auto* Dst  = GetDest<uint64_t*>(Data->SSAData, Node);
+  auto* Src1 = GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  auto* Src2 = GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+
+  const uint64_t TMP1 = (Selector & 0x01) == 0 ? Src1[0] : Src1[1];
+  const uint64_t TMP2 = (Selector & 0x10) == 0 ? Src2[0] : Src2[1];
+  
+  const auto make_lo = [](uint64_t lhs, uint64_t rhs) {
+    uint64_t result = 0;
+
+    for (size_t i = 0; i < 64; i++) {
+      if ((lhs & (1ULL << i)) != 0) {
+        result ^= rhs << i;
+      }
+    }
+
+    return result;
+  };
+  const auto make_hi = [](uint64_t lhs, uint64_t rhs) {
+    uint64_t result = 0;
+
+    for (size_t i = 1; i < 64; i++) {
+      if ((lhs & (1ULL << i)) != 0) {
+        result ^= rhs >> (64 - i);
+      }
+    }
+
+    return result;
+  };
+
+  Dst[0] = make_lo(TMP1, TMP2);
+  Dst[1] = make_hi(TMP1, TMP2);
+}
+
 #undef DEF_OP
 
 } // namespace FEXCore::CPU

External/FEXCore/Source/Interface/Core/Interpreter/F80Ops.cpp

@@ -356,6 +356,81 @@ DEF_OP(F80BCDSTORE) {
   memcpy(GDP, BCD, 10);
 }
 
+DEF_OP(F64SIN) {
+  auto Op = IROp->C<IR::IROp_F64SIN>();
+  double Src = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Tmp = sin(Src);
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64COS) {
+  auto Op = IROp->C<IR::IROp_F64COS>();
+  double Src = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Tmp = cos(Src);
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64TAN) {
+  auto Op = IROp->C<IR::IROp_F64TAN>();
+  double Src = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Tmp = tan(Src);
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64F2XM1) {
+  auto Op = IROp->C<IR::IROp_F64F2XM1>();
+  double Src = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Tmp = exp2(Src) - 1.0;
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64ATAN) {
+  auto Op = IROp->C<IR::IROp_F64ATAN>();
+  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
+  double Tmp = atan2(Src1, Src2);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64FPREM) {
+  auto Op = IROp->C<IR::IROp_F64FPREM>();
+  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
+  double Tmp = fmod(Src1, Src2);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64FPREM1) {
+  auto Op = IROp->C<IR::IROp_F64FPREM1>();
+  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
+  double Tmp = remainder(Src1, Src2);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64FYL2X) {
+  auto Op = IROp->C<IR::IROp_F64FYL2X>();
+  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
+  double Tmp = Src2 * log2(Src1);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64SCALE) {
+  auto Op = IROp->C<IR::IROp_F64SCALE>();
+  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
+  double trunc = (double)(int64_t)(Src2); //truncate
+  double Tmp = Src1 * exp2(trunc);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+
 #undef DEF_OP
 
 } // namespace FEXCore::CPU

External/FEXCore/Source/Interface/Core/Interpreter/F80Ops.h

@@ -222,6 +222,73 @@ struct OpHandlers<IR::OP_F80SCALE> {
   }
 };
 
+template<>
+struct OpHandlers<IR::OP_F64SIN> {
+  static double handle(double src) {
+    return sin(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64COS> {
+  static double handle(double src) {
+    return cos(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64TAN> {
+  static double handle(double src) {
+    return tan(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64F2XM1> {
+  static double handle(double src) {
+    return exp2(src) - 1.0;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64ATAN> {
+  static double handle(double src1, double src2) {
+    return atan2(src1, src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64FPREM> {
+  static double handle(double src1, double src2) {
+    return fmod(src1, src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64FPREM1> {
+  static double handle(double src1, double src2) {
+    return remainder(src1, src2);
+  }
+};
+
+
+template<>
+struct OpHandlers<IR::OP_F64FYL2X> {
+  static double handle(double src1, double src2) {
+    return src2 * log2(src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64SCALE> {
+  static double handle(double src1, double src2) {
+    double trunc = (double)(int64_t)(src2); //truncate
+    return src1 * exp2(trunc);
+  }
+};
+
+
+
 template<>
 struct OpHandlers<IR::OP_F80BCDSTORE> {
   static X80SoftFloat handle(X80SoftFloat Src1) {

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterClass.h

@@ -8,6 +8,7 @@
 #include <FEXCore/IR/IntrusiveIRList.h>
 
 namespace FEXCore::CPU {
+class Dispatcher;
 class X86DispatchGenerator;
 class Arm64DispatchGenerator;
 
@@ -20,28 +21,27 @@ using DestMapType = std::vector<uint32_t>;
 
 class InterpreterCore final : public CPUBackend {
 public:
-  explicit InterpreterCore(FEXCore::Context::Context *ctx,
-                           FEXCore::Core::InternalThreadState *Thread,
-                           bool CompileThread);
+  explicit InterpreterCore(Dispatcher *Dispatch,
+                           FEXCore::Core::InternalThreadState *Thread);
 
   [[nodiscard]] std::string GetName() override { return "Interpreter"; }
 
   [[nodiscard]] void *CompileCode(uint64_t Entry,
                                   FEXCore::IR::IRListView const *IR,
                                   FEXCore::Core::DebugData *DebugData,
-                                  FEXCore::IR::RegisterAllocationData *RAData) override;
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;
 
   [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }
 
   [[nodiscard]] bool NeedsOpDispatch() override { return true; }
 
-  void CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread);
+  static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
+  
+  void ClearCache() override;
 
 private:
-  FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *State;
-
-  std::unique_ptr<Dispatcher> Dispatcher{};
+  size_t BufferUsed;
+  Dispatcher *Dispatch;
 };
 
 template<typename T>

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterCore.cpp

@@ -9,66 +9,101 @@
 #include <FEXCore/Core/SignalDelegator.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/MathUtils.h>
 
 #include <memory>
 #include <signal.h>
 #include <stdint.h>
-#include <unordered_map>
 #include <utility>
 
 #include "InterpreterOps.h"
 
+#if defined(_M_X86_64)
+  #include "Interface/Core/Dispatcher/X86Dispatcher.h"
+#elif defined(_M_ARM_64)
+  #include "Interface/Core/Dispatcher/Arm64Dispatcher.h"
+#else
+  #error missing arch
+#endif
+
+static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
+static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 128;
+
 namespace FEXCore::IR {
   class IRListView;
   class RegisterAllocationData;
 }
 
+
 namespace FEXCore::CPU {
-class CPUBackend;
 
-static void InterpreterExecution(FEXCore::Core::CpuStateFrame *Frame) {
-  auto Thread = Frame->Thread;
+InterpreterCore::InterpreterCore(Dispatcher *Dispatcher, FEXCore::Core::InternalThreadState *Thread)
+  : CPUBackend(Thread, INITIAL_CODE_SIZE, MAX_CODE_SIZE)
+  , Dispatch(Dispatcher)
+  {
 
-  auto LocalEntry = Thread->LocalIRCache.find(Thread->CurrentFrame->State.rip);
+  auto &Interpreter = Thread->CurrentFrame->Pointers.Interpreter;
 
-  InterpreterOps::InterpretIR(Thread, Thread->CurrentFrame->State.rip, LocalEntry->second.IR.get(), LocalEntry->second.DebugData.get());
+  Interpreter.FragmentExecuter = reinterpret_cast<uint64_t>(&InterpreterOps::InterpretIR);
+
+  ClearCache();
 }
 
-InterpreterCore::InterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread)
-  : CTX {ctx}
-  , State {Thread} {
-
-  if (!CompileThread &&
-      CTX->Config.Core == FEXCore::Config::CONFIG_INTERPRETER) {
-    CreateAsmDispatch(ctx, Thread);
-    CTX->SignalDelegation->RegisterHostSignalHandler(SignalDelegator::SIGNAL_FOR_PAUSE, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      InterpreterCore *Core = reinterpret_cast<InterpreterCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleSignalPause(Signal, info, ucontext);
-    }, true);
+void InterpreterCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
 
 #ifdef _M_ARM_64
-    CTX->SignalDelegation->RegisterHostSignalHandler(SIGBUS, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      return FEXCore::ArchHelpers::Arm64::HandleSIGBUS(true, Signal, info, ucontext);
-    }, true);
+  CTX->SignalDelegation->RegisterHostSignalHandler(SIGBUS, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
+    return FEXCore::ArchHelpers::Arm64::HandleSIGBUS(true, Signal, info, ucontext);
+  }, true);
 #endif
+}
 
-    auto GuestSignalHandler = [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) -> bool {
-      InterpreterCore *Core = reinterpret_cast<InterpreterCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleGuestSignal(Signal, info, ucontext, GuestAction, GuestStack);
-    };
+void *InterpreterCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) {
 
-    for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
-      CTX->SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
-    }
+  const auto IRSize = AlignUp(IR->GetInlineSize(), 16);
+  const auto MaxSize = IRSize + Dispatcher::MaxInterpreterTrampolineSize + GDBEnabled * Dispatcher::MaxGDBPauseCheckSize;
+
+  if ((BufferUsed + MaxSize) > CurrentCodeBuffer->Size) {
+    ThreadState->CTX->ClearCodeCache(ThreadState);
   }
+
+  const auto BufferStart = CurrentCodeBuffer->Ptr + BufferUsed;
+
+  auto DestBuffer = BufferStart;
+
+  if (GDBEnabled) {
+    const auto GDBSize = Dispatch->GenerateGDBPauseCheck(DestBuffer, Entry);
+    DestBuffer += GDBSize;
+    BufferUsed += GDBSize;
+  }
+
+  const auto TrampolineSize = Dispatch->GenerateInterpreterTrampoline(DestBuffer);
+  DestBuffer += TrampolineSize;
+  BufferUsed += TrampolineSize;
+
+
+  IR->Serialize(DestBuffer);
+  DestBuffer += IRSize;
+  BufferUsed += IRSize;
+
+  return BufferStart;
 }
 
-void *InterpreterCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData) {
-  return reinterpret_cast<void*>(InterpreterExecution);
+void InterpreterCore::ClearCache() {
+  // Calling this one is needed to setup the initial CurrentCodeBuffer
+  [[maybe_unused]] auto CodeBuffer = GetEmptyCodeBuffer();
+  BufferUsed = 0;
 }
 
-std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread) {
-  return std::make_unique<InterpreterCore>(ctx, Thread, CompileThread);
+std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
+  return std::make_unique<InterpreterCore>(ctx->Dispatcher.get(), Thread);
 }
 
+void InitializeInterpreterSignalHandlers(FEXCore::Context::Context *CTX) {
+  InterpreterCore::InitializeSignalHandlers(CTX);
+}
+
+CPUBackendFeatures GetInterpreterBackendFeatures() {
+  return CPUBackendFeatures { };
+}
 }

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterCore.h

@@ -12,9 +12,11 @@ namespace FEXCore::Core {
 
 namespace FEXCore::CPU {
 class CPUBackend;
+struct DispatcherConfig;
 
 [[nodiscard]] std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx,
-                                                                FEXCore::Core::InternalThreadState *Thread,
-                                                                bool CompileThread);
+                                                                FEXCore::Core::InternalThreadState *Thread);
+void InitializeInterpreterSignalHandlers(FEXCore::Context::Context *CTX);
+CPUBackendFeatures GetInterpreterBackendFeatures();
 
 } // namespace FEXCore::CPU

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterFallbacks.cpp

@@ -47,6 +47,16 @@ FallbackInfo GetFallbackInfo(double(*fn)(X80SoftFloat), FEXCore::Core::FallbackH
   return {FABI_F64_F80, (void*)fn, HandlerIndex};
 }
 
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F64, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(double,double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F64_F64, (void*)fn, HandlerIndex};
+}
+
 template<>
 FallbackInfo GetFallbackInfo(int16_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
   return {FABI_I16_F80, (void*)fn, HandlerIndex};
@@ -122,6 +132,18 @@ void InterpreterOps::FillFallbackIndexPointers(uint64_t *Info) {
   Info[Core::OPINDEX_F80FPREM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FPREM1>::handle, Core::OPINDEX_F80FPREM1).fn);
   Info[Core::OPINDEX_F80FPREM] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FPREM>::handle, Core::OPINDEX_F80FPREM).fn);
   Info[Core::OPINDEX_F80SCALE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SCALE>::handle, Core::OPINDEX_F80SCALE).fn);
+
+  // Double Precision
+  Info[Core::OPINDEX_F64SIN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64SIN>::handle, Core::OPINDEX_F64SIN).fn);
+  Info[Core::OPINDEX_F64COS] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64COS>::handle, Core::OPINDEX_F64COS).fn);
+  Info[Core::OPINDEX_F64TAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64TAN>::handle, Core::OPINDEX_F64TAN).fn);
+  Info[Core::OPINDEX_F64ATAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64ATAN>::handle, Core::OPINDEX_F64ATAN).fn);
+  Info[Core::OPINDEX_F64F2XM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64F2XM1>::handle, Core::OPINDEX_F64F2XM1).fn);
+  Info[Core::OPINDEX_F64FYL2X] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FYL2X>::handle, Core::OPINDEX_F64FYL2X).fn);
+  Info[Core::OPINDEX_F64FPREM] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FPREM>::handle, Core::OPINDEX_F64FPREM).fn);
+  Info[Core::OPINDEX_F64FPREM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FPREM1>::handle, Core::OPINDEX_F64FPREM1).fn);
+  Info[Core::OPINDEX_F64SCALE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64SCALE>::handle, Core::OPINDEX_F64SCALE).fn);
+
 }
 
 bool InterpreterOps::GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info) {
@@ -238,6 +260,12 @@ bool InterpreterOps::GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Inf
       return true; \
     }
 
+#define COMMON_F64_OP(OP) \
+    case IR::OP_F64##OP: { \
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64##OP>::handle, Core::OPINDEX_F64##OP); \
+      return true; \
+    }
+
     // Unary
     COMMON_X87_OP(ROUND)
     COMMON_X87_OP(F2XM1)
@@ -261,6 +289,19 @@ bool InterpreterOps::GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Inf
     COMMON_X87_OP(FPREM)
     COMMON_X87_OP(SCALE)
 
+    // Double Precision Unary
+    COMMON_F64_OP(F2XM1)
+    COMMON_F64_OP(TAN)
+    COMMON_F64_OP(SIN)
+    COMMON_F64_OP(COS)
+
+    // Double Precision Binary
+    COMMON_F64_OP(FYL2X)
+    COMMON_F64_OP(ATAN)
+    COMMON_F64_OP(FPREM1)
+    COMMON_F64_OP(FPREM)
+    COMMON_F64_OP(SCALE)
+
     default:
       break;
   }

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.cpp

@@ -112,8 +112,6 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(ATOMICFETCHNEG,         AtomicFetchNeg);
 
   // Branch ops
-  REGISTER_OP(GUESTCALLDIRECT,        GuestCallDirect);
-  REGISTER_OP(GUESTCALLINDIRECT,      GuestCallIndirect);
   REGISTER_OP(SIGNALRETURN,           SignalReturn);
   REGISTER_OP(CALLBACKRETURN,         CallbackReturn);
   REGISTER_OP(EXITFUNCTION,           ExitFunction);
@@ -123,7 +121,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(INLINESYSCALL,          InlineSyscall);
   REGISTER_OP(THUNK,                  Thunk);
   REGISTER_OP(VALIDATECODE,           ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,        RemoveCodeEntry);
+  REGISTER_OP(REMOVETHREADCODEENTRY,        RemoveThreadCodeEntry);
   REGISTER_OP(CPUID,                  CPUID);
 
   // Conversion ops
@@ -166,6 +164,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(CODEBLOCK,              NoOp);
   REGISTER_OP(BEGINBLOCK,             NoOp);
   REGISTER_OP(ENDBLOCK,               NoOp);
+  REGISTER_OP(GUESTOPCODE,            NoOp);
   REGISTER_OP(FENCE,                  Fence);
   REGISTER_OP(BREAK,                  Break);
   REGISTER_OP(PHI,                    NoOp);
@@ -176,6 +175,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(INVALIDATEFLAGS,        NoOp);
   REGISTER_OP(PROCESSORID,            ProcessorID);
   REGISTER_OP(RDRAND,                 RDRAND);
+  REGISTER_OP(YIELD,                  Yield);
 
   // Move ops
   REGISTER_OP(EXTRACTELEMENTPAIR,     ExtractElementPair);
@@ -283,6 +283,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(VAESDECLAST,            AESDecLast);
   REGISTER_OP(VAESKEYGENASSIST,       AESKeyGenAssist);
   REGISTER_OP(CRC32,                  CRC32);
+  REGISTER_OP(PCLMUL,                 PCLMUL);
 
   // F80 ops
   REGISTER_OP(F80LOADFCW,             F80LOADFCW);
@@ -311,6 +312,17 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(F80BCDLOAD,             F80BCDLOAD);
   REGISTER_OP(F80BCDSTORE,            F80BCDSTORE);
 
+  // F64 ops
+  REGISTER_OP(F64SIN,                 F64SIN);
+  REGISTER_OP(F64COS,                 F64COS);
+  REGISTER_OP(F64TAN,                 F64TAN);
+  REGISTER_OP(F64F2XM1,               F64F2XM1);
+  REGISTER_OP(F64ATAN,                F64ATAN);
+  REGISTER_OP(F64FPREM,               F64FPREM);
+  REGISTER_OP(F64FPREM1,              F64FPREM1);
+  REGISTER_OP(F64FYL2X,               F64FYL2X);
+  REGISTER_OP(F64SCALE,               F64SCALE);
+
   return Handlers;
 }();
 
@@ -321,15 +333,9 @@ void InterpreterOps::Op_Unhandled(FEXCore::IR::IROp_Header *IROp, IROpData *Data
 void InterpreterOps::Op_NoOp(FEXCore::IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node) {
 }
 
-void InterpreterOps::InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData) {
+void InterpreterOps::InterpretIR(FEXCore::Core::CpuStateFrame *Frame, FEXCore::IR::IRListView const *CurrentIR) {
   volatile void *StackEntry = alloca(0);
 
-  // Debug data is only passed in debug builds
-  #ifndef NDEBUG
-  // TODO: should be moved to an IR Op
-  Thread->Stats.InstructionsExecuted.fetch_add(DebugData->GuestInstructionCount);
-  #endif
-
   uintptr_t ListSize = CurrentIR->GetSSACount();
 
   static_assert(sizeof(FEXCore::IR::IROp_Header) == 4);
@@ -338,9 +344,9 @@ void InterpreterOps::InterpretIR(FEXCore::Core::InternalThreadState *Thread, uin
   auto BlockEnd = CurrentIR->GetBlocks().end();
 
   InterpreterOps::IROpData OpData{};
-  OpData.State = Thread;
+  OpData.State = Frame->Thread;
   OpData.SSAData = alloca(ListSize * 16);
-  OpData.CurrentEntry = Entry;
+  OpData.CurrentEntry = Frame->State.rip;
   OpData.CurrentIR = CurrentIR;
   OpData.StackEntry = StackEntry;
   OpData.BlockIterator = CurrentIR->GetBlocks().begin();

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.h

@@ -28,6 +28,8 @@ namespace FEXCore::CPU {
     FABI_F80_I32,
     FABI_F32_F80,
     FABI_F64_F80,
+    FABI_F64_F64,
+    FABI_F64_F64_F64,
     FABI_I16_F80,
     FABI_I32_F80,
     FABI_I64_F80,
@@ -45,14 +47,14 @@ namespace FEXCore::CPU {
   class InterpreterOps {
 
     public:
-      static void InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData);
+      static void InterpretIR(FEXCore::Core::CpuStateFrame *Frame, FEXCore::IR::IRListView const *IR);
       static void FillFallbackIndexPointers(uint64_t *Info);
       static bool GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info);
 
       struct IROpData {
         FEXCore::Core::InternalThreadState *State{};
         uint64_t CurrentEntry{};
-        FEXCore::IR::IRListView *CurrentIR{};
+        FEXCore::IR::IRListView const *CurrentIR{};
         volatile void *StackEntry{};
         void *SSAData{};
         struct {
@@ -140,8 +142,6 @@ namespace FEXCore::CPU {
   DEF_OP(AtomicFetchNeg);
 
   ///< Branch ops
-  DEF_OP(GuestCallDirect);
-  DEF_OP(GuestCallIndirect);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -151,7 +151,7 @@ namespace FEXCore::CPU {
   DEF_OP(InlineSyscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveCodeEntry);
+  DEF_OP(RemoveThreadCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops
@@ -197,6 +197,7 @@ namespace FEXCore::CPU {
   DEF_OP(SetRoundingMode);
   DEF_OP(ProcessorID);
   DEF_OP(RDRAND);
+  DEF_OP(Yield);
 
   ///< Move ops
   DEF_OP(ExtractElementPair);
@@ -301,6 +302,7 @@ namespace FEXCore::CPU {
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
   DEF_OP(CRC32);
+  DEF_OP(PCLMUL);
 
   ///< F80 ops
   DEF_OP(F80LOADFCW);
@@ -328,6 +330,17 @@ namespace FEXCore::CPU {
   DEF_OP(F80CMP);
   DEF_OP(F80BCDLOAD);
   DEF_OP(F80BCDSTORE);
+
+  //< F64 ops
+  DEF_OP(F64SIN);
+  DEF_OP(F64COS);
+  DEF_OP(F64TAN);
+  DEF_OP(F64F2XM1);
+  DEF_OP(F64ATAN);
+  DEF_OP(F64FPREM);
+  DEF_OP(F64FPREM1);
+  DEF_OP(F64FYL2X);
+  DEF_OP(F64SCALE);
 #undef DEF_OP
   template<typename unsigned_type, typename signed_type, typename float_type>
   [[nodiscard]] static bool IsConditionTrue(uint8_t Cond, uint64_t Src1, uint64_t Src2) {
@@ -394,7 +407,7 @@ namespace FEXCore::CPU {
     return CompResult;
   }
 
-  static uint8_t GetOpSize(FEXCore::IR::IRListView *CurrentIR, IR::OrderedNodeWrapper Node) {
+  static uint8_t GetOpSize(FEXCore::IR::IRListView const *CurrentIR, IR::OrderedNodeWrapper Node) {
     auto IROp = CurrentIR->GetOp<FEXCore::IR::IROp_Header>(Node);
     return IROp->Size;
   }

External/FEXCore/Source/Interface/Core/Interpreter/MemoryOps.cpp

@@ -4,6 +4,7 @@ tags: backend|interpreter
 $end_info$
 */
 
+#include "Interface/Core/CPUID.h"
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"

External/FEXCore/Source/Interface/Core/Interpreter/MiscOps.cpp

@@ -4,6 +4,8 @@ tags: backend|interpreter
 $end_info$
 */
 
+#include "Interface/Context/Context.h"
+
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
@@ -157,6 +159,11 @@ DEF_OP(RDRAND) {
   // Second result is if we managed to read a valid random number or not
   DstPtr[1] = Result == 8 ? 1 : 0;
 }
+
+DEF_OP(Yield) {
+  // Nop implementation
+}
+
 #undef DEF_OP
 
 } // namespace FEXCore::CPU

External/FEXCore/Source/Interface/Core/JIT/Arm64/Arm64Relocations.cpp

@@ -0,0 +1,131 @@
+/*
+$info$
+tags: backend|arm64
+desc: relocation logic of the arm64 splatter backend
+$end_info$
+*/
+#include "Interface/Context/Context.h"
+#include "Interface/Core/JIT/Arm64/JITClass.h"
+#include "Interface/HLE/Thunks/Thunks.h"
+
+namespace FEXCore::CPU {
+    
+uint64_t Arm64JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+  switch (Op) {
+    case FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER:
+      return ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker;
+    break;
+    default:
+      ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
+    break;
+  }
+  return ~0ULL;
+}
+
+void Arm64JITCore::InsertNamedThunkRelocation(vixl::aarch64::Register Reg, const IR::SHA256Sum &Sum) {
+  Relocation MoveABI{};
+  MoveABI.NamedThunkMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE;
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = GetCursorAddress<uint8_t *>();
+  MoveABI.NamedThunkMove.Offset = CurrentCursor - GuestEntry;
+  MoveABI.NamedThunkMove.Symbol = Sum;
+  MoveABI.NamedThunkMove.RegisterIndex = Reg.GetCode();
+
+  uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Sum));
+
+  LoadConstant(Reg, Pointer, EmitterCTX->Config.CacheObjectCodeCompilation());
+  Relocations.emplace_back(MoveABI);
+}
+
+Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+  uint64_t Pointer = GetNamedSymbolLiteral(Op);
+
+  Arm64JITCore::NamedSymbolLiteralPair Lit {
+    .Lit = Literal(Pointer),
+    .MoveABI = {
+      .NamedSymbolLiteral = {
+        .Header = {
+          .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL,
+        },
+        .Symbol = Op,
+        .Offset = 0,
+      },
+    },
+  };
+  return Lit;
+}
+
+void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit) {
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = GetCursorAddress<uint8_t *>();
+  Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - GuestEntry;
+
+  place(&Lit.Lit);
+  Relocations.emplace_back(Lit.MoveABI);
+}
+
+void Arm64JITCore::InsertGuestRIPMove(vixl::aarch64::Register Reg, uint64_t Constant) {
+  Relocation MoveABI{};
+  MoveABI.GuestRIPMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE;
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = GetCursorAddress<uint8_t *>();
+  MoveABI.GuestRIPMove.Offset = CurrentCursor - GuestEntry;
+  MoveABI.GuestRIPMove.GuestRIP = Constant;
+  MoveABI.GuestRIPMove.RegisterIndex = Reg.GetCode();
+
+  LoadConstant(Reg, Constant, EmitterCTX->Config.CacheObjectCodeCompilation());
+  Relocations.emplace_back(MoveABI);
+}
+
+bool Arm64JITCore::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations) {
+  size_t DataIndex{};
+  for (size_t j = 0; j < NumRelocations; ++j) {
+    const FEXCore::CPU::Relocation *Reloc = reinterpret_cast<const FEXCore::CPU::Relocation *>(&EntryRelocations[DataIndex]);
+    LOGMAN_THROW_A_FMT((DataIndex % alignof(Relocation)) == 0, "Alignment of relocation wasn't adhered to");
+
+    switch (Reloc->Header.Type) {
+      case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
+        uint64_t Pointer = GetNamedSymbolLiteral(Reloc->NamedSymbolLiteral.Symbol);
+        // Relocation occurs at the cursorEntry + offset relative to that cursor
+        GetBuffer()->SetCursorOffset(CursorEntry + Reloc->NamedSymbolLiteral.Offset);
+
+        // Generate a literal so we can place it
+        Literal<uint64_t> Lit(Pointer);
+        place(&Lit);
+
+        DataIndex += sizeof(Reloc->NamedSymbolLiteral);
+        break;
+      }
+      case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
+        uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Reloc->NamedThunkMove.Symbol));
+        if (Pointer == ~0ULL) {
+          return false;
+        }
+
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        GetBuffer()->SetCursorOffset(CursorEntry + Reloc->NamedThunkMove.Offset);
+        LoadConstant(vixl::aarch64::XRegister(Reloc->NamedThunkMove.RegisterIndex), Pointer, true);
+        DataIndex += sizeof(Reloc->NamedThunkMove);
+        break;
+      }
+      case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE: {
+        // XXX: Reenable once the JIT Object Cache is upstream
+        // XXX: Should spin the relocation list, create a list of guest RIP moves, and ask for them all once, reduces lock contention.
+        uint64_t Pointer = ~0ULL; // EmitterCTX->JITObjectCache->FindRelocatedRIP(Reloc->GuestRIPMove.GuestRIP);
+        if (Pointer == ~0ULL) {
+          return false;
+        }
+
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        GetBuffer()->SetCursorOffset(CursorEntry + Reloc->GuestRIPMove.Offset);
+        LoadConstant(vixl::aarch64::XRegister(Reloc->GuestRIPMove.RegisterIndex), Pointer, true);
+        DataIndex += sizeof(Reloc->GuestRIPMove);
+        break;
+      }
+    }
+  }
+
+  return true;
+}
+}
+

External/FEXCore/Source/Interface/Core/JIT/Arm64/AtomicOps.cpp

@@ -4,6 +4,7 @@ tags: backend|arm64
 $end_info$
 */
 
+#include "Interface/Context/Context.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 
 namespace FEXCore::CPU {
@@ -509,10 +510,10 @@ DEF_OP(AtomicSwap) {
   if (CTX->HostFeatures.SupportsAtomics) {
     mov(TMP2, GetReg<RA_64>(Op->Value.ID()));
     switch (IROp->Size) {
-    case 1: swplb(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 2: swplh(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 4: swpl(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 8: swpl(TMP2.X(), GetReg<RA_64>(Node), MemOperand(MemSrc)); break;
+    case 1: swpalb(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 2: swpalh(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 4: swpal(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 8: swpal(TMP2.X(), GetReg<RA_64>(Node), MemOperand(MemSrc)); break;
     default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
     }
   }

External/FEXCore/Source/Interface/Core/JIT/Arm64/BranchOps.cpp

@@ -4,6 +4,7 @@ tags: backend|arm64
 $end_info$
 */
 
+#include "Interface/Context/Context.h"
 #include "FEXCore/IR/IR.h"
 #include "Interface/Core/LookupCache.h"
 
@@ -19,13 +20,6 @@ namespace FEXCore::CPU {
 using namespace vixl;
 using namespace vixl::aarch64;
 #define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
 
 DEF_OP(SignalReturn) {
   // First we must reset the stack
@@ -33,7 +27,7 @@ DEF_OP(SignalReturn) {
 
   // Now branch to our signal return helper
   // This can't be a direct branch since the code needs to live at a constant location
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SignalReturnHandler)));
+  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SignalReturnHandler)));
   br(x0);
 }
 
@@ -46,10 +40,10 @@ DEF_OP(CallbackReturn) {
   ResetStack();
 
   // We can now lower the ref counter again
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SignalHandlerRefCountPointer)));
-  ldr(w2, MemOperand(x0));
+  
+  ldr(w2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter)));
   sub(w2, w2, 1);
-  str(w2, MemOperand(x0));
+  str(w2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter)));
 
   // We need to adjust an additional 8 bytes to get back to the original "misaligned" RSP state
   ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
@@ -73,7 +67,7 @@ DEF_OP(ExitFunction) {
   uint64_t NewRIP;
 
   if (IsInlineConstant(Op->NewRIP, &NewRIP) || IsInlineEntrypointOffset(Op->NewRIP, &NewRIP)) {
-    Literal l_BranchHost{ThreadSharedData.Dispatcher->ExitFunctionLinkerAddress};
+    Literal l_BranchHost{ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker};
     Literal l_BranchGuest{NewRIP};
 
     ldr(x0, &l_BranchHost);
@@ -82,10 +76,10 @@ DEF_OP(ExitFunction) {
     place(&l_BranchHost);
     place(&l_BranchGuest);
   } else {
-    RipReg = GetReg<RA_64>(Op->Header.Args[0].ID());
+    RipReg = GetReg<RA_64>(Op->NewRIP.ID());
 
     // L1 Cache
-    ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.L1Pointer)));
+    ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer)));
 
     and_(x3, RipReg, LookupCache::L1_ENTRIES_MASK);
     add(x0, x0, Operand(x3, Shift::LSL, 4));
@@ -96,7 +90,7 @@ DEF_OP(ExitFunction) {
     br(x1);
 
     bind(&FullLookup);
-    ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.DispatcherLoopTop)));
+    ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop)));
     str(RipReg, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
     br(TMP1);
   }
@@ -104,9 +98,9 @@ DEF_OP(ExitFunction) {
 
 DEF_OP(Jump) {
   const auto Op = IROp->C<IR::IROp_Jump>();
-  const auto ArgID = Op->Args(0).ID();
+  const auto Target = Op->TargetBlock.ID();
 
-  PendingTargetLabel = &JumpTargets.try_emplace(ArgID).first->second;
+  PendingTargetLabel = &JumpTargets.try_emplace(Target).first->second;
 }
 
 #define GRCMP(Node) (Op->CompareSize == 4 ? GetReg<RA_32>(Node) : GetReg<RA_64>(Node))
@@ -199,8 +193,8 @@ DEF_OP(Syscall) {
     str(GetReg<RA_64>(Op->Header.Args[i].ID()), MemOperand(sp, i * 8));
   }
 
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SyscallHandlerObj)));
-  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SyscallHandlerFunc)));
+  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerObj)));
+  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerFunc)));
   mov(x1, STATE);
   mov(x2, sp);
   blr(x3);
@@ -383,7 +377,7 @@ DEF_OP(Thunk) {
 
   PushDynamicRegsAndLR();
 
-  mov(x0, GetReg<RA_64>(Op->Header.Args[0].ID()));
+  mov(x0, GetReg<RA_64>(Op->ArgPtr.ID()));
 
   auto thunkFn = ThreadState->CTX->ThunkHandler->LookupThunk(Op->ThunkNameHash);
   LoadConstant(x2, (uintptr_t)thunkFn);
@@ -442,7 +436,7 @@ DEF_OP(ValidateCode) {
   }
 }
 
-DEF_OP(RemoveCodeEntry) {
+DEF_OP(RemoveThreadCodeEntry) {
   // Arguments are passed as follows:
   // X0: Thread
   // X1: RIP
@@ -452,7 +446,7 @@ DEF_OP(RemoveCodeEntry) {
   mov(x0, STATE);
   LoadConstant(x1, Entry);
 
-  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.RemoveCodeEntryFromJIT)));
+  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.RemoveThreadCodeEntryFromJIT)));
   SpillStaticRegs();
   blr(x2);
   FillStaticRegs();
@@ -469,10 +463,10 @@ DEF_OP(CPUID) {
   // x0 = CPUID Handler
   // x1 = CPUID Function
   // x2 = CPUID Leaf
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.CPUIDObj)));
-  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.CPUIDFunction)));
-  mov(x1, GetReg<RA_64>(Op->Header.Args[0].ID()));
-  mov(x2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj)));
+  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDFunction)));
+  mov(x1, GetReg<RA_64>(Op->Function.ID()));
+  mov(x2, GetReg<RA_64>(Op->Leaf.ID()));
   SpillStaticRegs();
   blr(x3);
   FillStaticRegs();
@@ -489,8 +483,6 @@ DEF_OP(CPUID) {
 #undef DEF_OP
 void Arm64JITCore::RegisterBranchHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(GUESTCALLDIRECT,   GuestCallDirect);
-  REGISTER_OP(GUESTCALLINDIRECT, GuestCallIndirect);
   REGISTER_OP(SIGNALRETURN,      SignalReturn);
   REGISTER_OP(CALLBACKRETURN,    CallbackReturn);
   REGISTER_OP(EXITFUNCTION,      ExitFunction);
@@ -500,7 +492,7 @@ void Arm64JITCore::RegisterBranchHandlers() {
   REGISTER_OP(INLINESYSCALL,     InlineSyscall);
   REGISTER_OP(THUNK,             Thunk);
   REGISTER_OP(VALIDATECODE,      ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,   RemoveCodeEntry);
+  REGISTER_OP(REMOVETHREADCODEENTRY,   RemoveThreadCodeEntry);
   REGISTER_OP(CPUID,             CPUID);
 #undef REGISTER_OP
 }

External/FEXCore/Source/Interface/Core/JIT/Arm64/ConversionOps.cpp

@@ -13,22 +13,22 @@ using namespace vixl::aarch64;
 #define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 DEF_OP(VInsGPR) {
   auto Op = IROp->C<IR::IROp_VInsGPR>();
-  mov(GetDst(Node), GetSrc(Op->Header.Args[0].ID()));
+  mov(GetDst(Node), GetSrc(Op->DestVector.ID()));
   switch (Op->Header.ElementSize) {
     case 1: {
-      ins(GetDst(Node).V16B(), Op->DestIdx, GetReg<RA_32>(Op->Header.Args[1].ID()));
+      ins(GetDst(Node).V16B(), Op->DestIdx, GetReg<RA_32>(Op->Src.ID()));
     break;
     }
     case 2: {
-      ins(GetDst(Node).V8H(), Op->DestIdx, GetReg<RA_32>(Op->Header.Args[1].ID()));
+      ins(GetDst(Node).V8H(), Op->DestIdx, GetReg<RA_32>(Op->Src.ID()));
     break;
     }
     case 4: {
-      ins(GetDst(Node).V4S(), Op->DestIdx, GetReg<RA_32>(Op->Header.Args[1].ID()));
+      ins(GetDst(Node).V4S(), Op->DestIdx, GetReg<RA_32>(Op->Src.ID()));
     break;
     }
     case 8: {
-      ins(GetDst(Node).V2D(), Op->DestIdx, GetReg<RA_64>(Op->Header.Args[1].ID()));
+      ins(GetDst(Node).V2D(), Op->DestIdx, GetReg<RA_64>(Op->Src.ID()));
     break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
@@ -39,18 +39,18 @@ DEF_OP(VCastFromGPR) {
   auto Op = IROp->C<IR::IROp_VCastFromGPR>();
   switch (Op->Header.ElementSize) {
     case 1:
-      uxtb(TMP1.W(), GetReg<RA_32>(Op->Header.Args[0].ID()));
+      uxtb(TMP1.W(), GetReg<RA_32>(Op->Src.ID()));
       fmov(GetDst(Node).S(), TMP1.W());
       break;
     case 2:
-      uxth(TMP1.W(), GetReg<RA_32>(Op->Header.Args[0].ID()));
+      uxth(TMP1.W(), GetReg<RA_32>(Op->Src.ID()));
       fmov(GetDst(Node).S(), TMP1.W());
       break;
     case 4:
-      fmov(GetDst(Node).S(), GetReg<RA_32>(Op->Header.Args[0].ID()).W());
+      fmov(GetDst(Node).S(), GetReg<RA_32>(Op->Src.ID()).W());
       break;
     case 8:
-      fmov(GetDst(Node).D(), GetReg<RA_64>(Op->Header.Args[0].ID()).X());
+      fmov(GetDst(Node).D(), GetReg<RA_64>(Op->Src.ID()).X());
       break;
     default: LOGMAN_MSG_A_FMT("Unknown castGPR element size: {}", Op->Header.ElementSize);
   }
@@ -58,22 +58,22 @@ DEF_OP(VCastFromGPR) {
 
 DEF_OP(Float_FromGPR_S) {
   auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
   switch (Conv) {
     case 0x0404: { // Float <- int32_t
-      scvtf(GetDst(Node).S(), GetReg<RA_32>(Op->Header.Args[0].ID()));
+      scvtf(GetDst(Node).S(), GetReg<RA_32>(Op->Src.ID()));
       break;
     }
     case 0x0408: { // Float <- int64_t
-      scvtf(GetDst(Node).S(), GetReg<RA_64>(Op->Header.Args[0].ID()));
+      scvtf(GetDst(Node).S(), GetReg<RA_64>(Op->Src.ID()));
       break;
     }
     case 0x0804: { // Double <- int32_t
-      scvtf(GetDst(Node).D(), GetReg<RA_32>(Op->Header.Args[0].ID()));
+      scvtf(GetDst(Node).D(), GetReg<RA_32>(Op->Src.ID()));
       break;
     }
     case 0x0808: { // Double <- int64_t
-      scvtf(GetDst(Node).D(), GetReg<RA_64>(Op->Header.Args[0].ID()));
+      scvtf(GetDst(Node).D(), GetReg<RA_64>(Op->Src.ID()));
       break;
     }
   }
@@ -81,14 +81,14 @@ DEF_OP(Float_FromGPR_S) {
 
 DEF_OP(Float_FToF) {
   auto Op = IROp->C<IR::IROp_Float_FToF>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
   switch (Conv) {
     case 0x0804: { // Double <- Float
-      fcvt(GetDst(Node).D(), GetSrc(Op->Header.Args[0].ID()).S());
+      fcvt(GetDst(Node).D(), GetSrc(Op->Scalar.ID()).S());
       break;
     }
     case 0x0408: { // Float <- Double
-      fcvt(GetDst(Node).S(), GetSrc(Op->Header.Args[0].ID()).D());
+      fcvt(GetDst(Node).S(), GetSrc(Op->Scalar.ID()).D());
       break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown FCVT sizes: 0x{:x}", Conv);
@@ -99,10 +99,10 @@ DEF_OP(Vector_SToF) {
   auto Op = IROp->C<IR::IROp_Vector_SToF>();
   switch (Op->Header.ElementSize) {
     case 4:
-      scvtf(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+      scvtf(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
     break;
     case 8:
-      scvtf(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+      scvtf(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
     break;
     default: LOGMAN_MSG_A_FMT("Unknown Vector_SToF element size: {}", Op->Header.ElementSize);
   }
@@ -112,10 +112,10 @@ DEF_OP(Vector_FToZS) {
   auto Op = IROp->C<IR::IROp_Vector_FToZS>();
   switch (Op->Header.ElementSize) {
     case 4:
-      fcvtzs(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+      fcvtzs(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
     break;
     case 8:
-      fcvtzs(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+      fcvtzs(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
     break;
     default: LOGMAN_MSG_A_FMT("Unknown Vector_FToZS element size: {}", Op->Header.ElementSize);
   }
@@ -125,11 +125,11 @@ DEF_OP(Vector_FToS) {
   auto Op = IROp->C<IR::IROp_Vector_FToS>();
   switch (Op->Header.ElementSize) {
     case 4:
-      frinti(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+      frinti(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
       fcvtzs(GetDst(Node).V4S(), GetDst(Node).V4S());
     break;
     case 8:
-      frinti(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+      frinti(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
       fcvtzs(GetDst(Node).V2D(), GetDst(Node).V2D());
     break;
     default: LOGMAN_MSG_A_FMT("Unknown Vector_FToS element size: {}", Op->Header.ElementSize);
@@ -142,11 +142,11 @@ DEF_OP(Vector_FToF) {
 
   switch (Conv) {
     case 0x0804: { // Double <- Float
-      fcvtl(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2S());
+      fcvtl(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2S());
       break;
     }
     case 0x0408: { // Float <- Double
-      fcvtn(GetDst(Node).V2S(), GetSrc(Op->Header.Args[0].ID()).V2D());
+      fcvtn(GetDst(Node).V2S(), GetSrc(Op->Vector.ID()).V2D());
       break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown Vector_FToF Type : 0x{:04x}", Conv); break;
@@ -159,50 +159,50 @@ DEF_OP(Vector_FToI) {
     case FEXCore::IR::Round_Nearest.Val:
       switch (Op->Header.ElementSize) {
         case 4:
-          frintn(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+          frintn(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
         break;
         case 8:
-          frintn(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+          frintn(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
         break;
       }
     break;
     case FEXCore::IR::Round_Negative_Infinity.Val:
       switch (Op->Header.ElementSize) {
         case 4:
-          frintm(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+          frintm(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
         break;
         case 8:
-          frintm(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+          frintm(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
         break;
       }
     break;
     case FEXCore::IR::Round_Positive_Infinity.Val:
       switch (Op->Header.ElementSize) {
         case 4:
-          frintp(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+          frintp(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
         break;
         case 8:
-          frintp(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+          frintp(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
         break;
       }
     break;
     case FEXCore::IR::Round_Towards_Zero.Val:
       switch (Op->Header.ElementSize) {
         case 4:
-          frintz(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+          frintz(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
         break;
         case 8:
-          frintz(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+          frintz(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
         break;
       }
     break;
     case FEXCore::IR::Round_Host.Val:
       switch (Op->Header.ElementSize) {
         case 4:
-          frinti(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+          frinti(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
         break;
         case 8:
-          frinti(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+          frinti(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
         break;
       }
     break;

External/FEXCore/Source/Interface/Core/JIT/Arm64/EncryptionOps.cpp

@@ -14,41 +14,41 @@ using namespace vixl::aarch64;
 
 DEF_OP(AESImc) {
   auto Op = IROp->C<IR::IROp_VAESImc>();
-  aesimc(GetDst(Node).V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
+  aesimc(GetDst(Node).V16B(), GetSrc(Op->Vector.ID()).V16B());
 }
 
 DEF_OP(AESEnc) {
 	auto Op = IROp->C<IR::IROp_VAESEnc>();
   eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
+  mov(VTMP1.V16B(), GetSrc(Op->State.ID()).V16B());
   aese(VTMP1.V16B(), VTMP2.V16B());
   aesmc(VTMP1.V16B(), VTMP1.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].ID()).V16B());
+  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Key.ID()).V16B());
 }
 
 DEF_OP(AESEncLast) {
 	auto Op = IROp->C<IR::IROp_VAESEncLast>();
   eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
+  mov(VTMP1.V16B(), GetSrc(Op->State.ID()).V16B());
   aese(VTMP1.V16B(), VTMP2.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].ID()).V16B());
+  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Key.ID()).V16B());
 }
 
 DEF_OP(AESDec) {
 	auto Op = IROp->C<IR::IROp_VAESDec>();
   eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
+  mov(VTMP1.V16B(), GetSrc(Op->State.ID()).V16B());
   aesd(VTMP1.V16B(), VTMP2.V16B());
   aesimc(VTMP1.V16B(), VTMP1.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].ID()).V16B());
+  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Key.ID()).V16B());
 }
 
 DEF_OP(AESDecLast) {
 	auto Op = IROp->C<IR::IROp_VAESDecLast>();
   eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
+  mov(VTMP1.V16B(), GetSrc(Op->State.ID()).V16B());
   aesd(VTMP1.V16B(), VTMP2.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].ID()).V16B());
+  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Key.ID()).V16B());
 }
 
 DEF_OP(AESKeyGenAssist) {
@@ -59,7 +59,7 @@ DEF_OP(AESKeyGenAssist) {
 
   // Do a "regular" AESE step
   eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
+  mov(VTMP1.V16B(), GetSrc(Op->Src.ID()).V16B());
   aese(VTMP1.V16B(), VTMP2.V16B());
 
   // Do a table shuffle to undo ShiftRows
@@ -102,16 +102,45 @@ DEF_OP(CRC32) {
   }
 }
 
+DEF_OP(PCLMUL) {
+  auto Op = IROp->C<IR::IROp_PCLMUL>();
+
+  auto Dst  = GetDst(Node).Q();
+  auto Src1 = GetSrc(Op->Src1.ID()).V2D();
+  auto Src2 = GetSrc(Op->Src2.ID()).V2D();
+
+  switch (Op->Selector) {
+  case 0b00000000:
+    pmull(Dst, Src1, Src2);
+    break;
+  case 0b00000001:
+    mov(VTMP1.V1D(), Src1, 1);
+    pmull(Dst, VTMP1.V2D(), Src2);
+    break;
+  case 0b00010000:
+    mov(VTMP1.V1D(), Src2, 1);
+    pmull(Dst, VTMP1.V2D(), Src1);
+    break;
+  case 0b00010001:
+    pmull2(Dst, Src1, Src2);
+    break;
+  default:
+    LOGMAN_MSG_A_FMT("Unknown PCLMUL selector: {}", Op->Selector);
+    break;
+  }
+}
+
 #undef DEF_OP
 void Arm64JITCore::RegisterEncryptionHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(VAESIMC,     AESImc);
-  REGISTER_OP(VAESENC,     AESEnc);
-  REGISTER_OP(VAESENCLAST, AESEncLast);
-  REGISTER_OP(VAESDEC,     AESDec);
-  REGISTER_OP(VAESDECLAST, AESDecLast);
-  REGISTER_OP(VAESKEYGENASSIST, AESKeyGenAssist);
+  REGISTER_OP(VAESIMC,           AESImc);
+  REGISTER_OP(VAESENC,           AESEnc);
+  REGISTER_OP(VAESENCLAST,       AESEncLast);
+  REGISTER_OP(VAESDEC,           AESDec);
+  REGISTER_OP(VAESDECLAST,       AESDecLast);
+  REGISTER_OP(VAESKEYGENASSIST,  AESKeyGenAssist);
   REGISTER_OP(CRC32,             CRC32);
+  REGISTER_OP(PCLMUL,            PCLMUL);
 #undef REGISTER_OP
 }
 }

External/FEXCore/Source/Interface/Core/JIT/Arm64/FlagOps.cpp

@@ -13,7 +13,7 @@ using namespace vixl::aarch64;
 #define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 DEF_OP(GetHostFlag) {
   auto Op = IROp->C<IR::IROp_GetHostFlag>();
-  ubfx(GetReg<RA_64>(Node), GetReg<RA_64>(Op->Header.Args[0].ID()), Op->Flag, 1);
+  ubfx(GetReg<RA_64>(Node), GetReg<RA_64>(Op->Value.ID()), Op->Flag, 1);
 }
 
 #undef DEF_OP

External/FEXCore/Source/Interface/Core/JIT/Arm64/JIT.cpp

@@ -35,6 +35,10 @@ $end_info$
 #include <unistd.h>
 #include <string.h>
 
+static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
+// We don't want to move above 128MB atm because that means we will have to encode longer jumps
+static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 128;
+
 namespace {
 static uint64_t LUDIV(uint64_t SrcHigh, uint64_t SrcLow, uint64_t Divisor) {
   __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
@@ -71,11 +75,6 @@ static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
 
 namespace FEXCore::CPU {
 
-void Arm64JITCore::CopyNecessaryDataForCompileThread(CPUBackend *Original) {
-  Arm64JITCore *Core = reinterpret_cast<Arm64JITCore*>(Original);
-  ThreadSharedData = Core->ThreadSharedData;
-}
-
 using namespace vixl;
 using namespace vixl::aarch64;
 
@@ -93,7 +92,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushDynamicRegsAndLR();
 
         uxth(w0, GetReg<RA_32>(IROp->Args[0].ID()));
-        ldr(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x1);
 
         PopDynamicRegsAndLR();
@@ -108,7 +107,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushDynamicRegsAndLR();
 
         fmov(v0.S(), GetSrc(IROp->Args[0].ID()).S()) ;
-        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x0);
 
         PopDynamicRegsAndLR();
@@ -127,7 +126,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushDynamicRegsAndLR();
 
         mov(v0.D(), GetSrc(IROp->Args[0].ID()).D());
-        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x0);
 
         PopDynamicRegsAndLR();
@@ -152,7 +151,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         else {
           mov(w0, GetReg<RA_32>(IROp->Args[0].ID()));
         }
-        ldr(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x1);
 
         PopDynamicRegsAndLR();
@@ -173,7 +172,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
         umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -192,7 +191,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
         umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -203,6 +202,43 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
       }
       break;
 
+      case FABI_F64_F64: {
+        SpillStaticRegs();
+
+        PushDynamicRegsAndLR();
+
+        mov(v0.D(), GetSrc(IROp->Args[0].ID()).D());
+        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        blr(x0);
+
+        PopDynamicRegsAndLR();
+
+        FillStaticRegs();
+
+        mov(GetDst(Node).D(), v0.D());
+
+      }
+      break;
+
+      case FABI_F64_F64_F64: {
+        SpillStaticRegs();
+
+        PushDynamicRegsAndLR();
+
+        mov(v0.D(), GetSrc(IROp->Args[0].ID()).D());
+        mov(v1.D(), GetSrc(IROp->Args[1].ID()).D());
+        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        blr(x0);
+
+        PopDynamicRegsAndLR();
+
+        FillStaticRegs();
+
+        mov(GetDst(Node).D(), v0.D());
+
+      }
+      break;
+
       case FABI_I16_F80:{
         SpillStaticRegs();
 
@@ -211,7 +247,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
         umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -229,7 +265,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
         umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -247,7 +283,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
         umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -268,7 +304,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         umov(x2, GetSrc(IROp->Args[1].ID()).V2D(), 0);
         umov(w3, GetSrc(IROp->Args[1].ID()).V8H(), 4);
 
-        ldr(x4, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x4, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x4);
 
         PopDynamicRegsAndLR();
@@ -286,7 +322,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
         umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -309,7 +345,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         umov(x2, GetSrc(IROp->Args[1].ID()).V2D(), 0);
         umov(w3, GetSrc(IROp->Args[1].ID()).V8H(), 4);
 
-        ldr(x4, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(x4, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x4);
 
         PopDynamicRegsAndLR();
@@ -332,81 +368,63 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
   }
 }
 
+
+static uint64_t Arm64JITCore_ExitFunctionLink(FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
+  auto Thread = Frame->Thread;
+  auto GuestRip = record[1];
+
+  auto HostCode = Thread->LookupCache->FindBlock(GuestRip);
+
+  if (!HostCode) {
+    //fmt::print("ExitFunctionLink: Aborting, {:X} not in cache\n", GuestRip);
+    Frame->State.rip = GuestRip;
+    return Frame->Pointers.Common.DispatcherLoopTop;
+  }
+
+  uintptr_t branch = (uintptr_t)(record) - 8;
+  auto LinkerAddress = Frame->Pointers.Common.ExitFunctionLinker;
+
+  auto offset = HostCode/4 - branch/4;
+  if (IsInt26(offset)) {
+    // optimal case - can branch directly
+    // patch the code
+    vixl::aarch64::Assembler emit((uint8_t*)(branch), 24);
+    vixl::CodeBufferCheckScope scope(&emit, 24, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
+    emit.b(offset);
+    emit.FinalizeCode();
+    vixl::aarch64::CPU::EnsureIAndDCacheCoherency((void*)branch, 24);
+
+    // Add de-linking handler
+    Thread->LookupCache->AddBlockLink(GuestRip, (uintptr_t)record, [branch, LinkerAddress]{
+      vixl::aarch64::Assembler emit((uint8_t*)(branch), 24);
+      vixl::CodeBufferCheckScope scope(&emit, 24, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
+      Literal l_BranchHost{LinkerAddress};
+      emit.ldr(x0, &l_BranchHost);
+      emit.blr(x0);
+      emit.place(&l_BranchHost);
+      emit.FinalizeCode();
+      vixl::aarch64::CPU::EnsureIAndDCacheCoherency((void*)branch, 24);
+    });
+  } else {
+    // fallback case - do a soft-er link by patching the pointer
+    record[0] = HostCode;
+
+    // Add de-linking handler
+    Thread->LookupCache->AddBlockLink(GuestRip, (uintptr_t)record, [record, LinkerAddress]{
+      record[0] = LinkerAddress;
+    });
+  }
+
+  return HostCode;
+}
+
 void Arm64JITCore::Op_NoOp(IR::IROp_Header *IROp, IR::NodeID Node) {
 }
 
-Arm64JITCore::CodeBuffer Arm64JITCore::AllocateNewCodeBuffer(size_t Size) {
-  CodeBuffer Buffer;
-  Buffer.Size = Size;
-  Buffer.Ptr = static_cast<uint8_t*>(
-               FEXCore::Allocator::mmap(nullptr,
-                    Buffer.Size,
-                    PROT_READ | PROT_WRITE | PROT_EXEC,
-                    MAP_PRIVATE | MAP_ANONYMOUS,
-                    -1, 0));
-  LOGMAN_THROW_A_FMT(!!Buffer.Ptr, "Couldn't allocate code buffer");
-  Dispatcher->RegisterCodeBuffer(Buffer.Ptr, Buffer.Size);
-  if (CTX->Config.GlobalJITNaming()) {
-    CTX->Symbols.RegisterJITSpace(Buffer.Ptr, Buffer.Size);
-  }
-  return Buffer;
-}
-
-void Arm64JITCore::FreeCodeBuffer(CodeBuffer Buffer) {
-  FEXCore::Allocator::munmap(Buffer.Ptr, Buffer.Size);
-  Dispatcher->RemoveCodeBuffer(Buffer.Ptr);
-}
-
-Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread)
-  : Arm64Emitter(ctx, 0)
-  , CTX {ctx}
-  , ThreadState {Thread} {
-
-  {
-    // Set up pointers that the JIT needs to load
-    auto &Pointers = ThreadState->CurrentFrame->Pointers.AArch64;
-    // Process specific
-    Pointers.LUDIV = reinterpret_cast<uint64_t>(LUDIV);
-    Pointers.LDIV = reinterpret_cast<uint64_t>(LDIV);
-    Pointers.LUREM = reinterpret_cast<uint64_t>(LUREM);
-    Pointers.LREM = reinterpret_cast<uint64_t>(LREM);
-    Pointers.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
-    Pointers.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
-    Pointers.RemoveCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveCodeEntryFromJit);
-    Pointers.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
-
-    {
-      FEXCore::Utils::MemberFunctionToPointerCast PMF(&FEXCore::CPUIDEmu::RunFunction);
-      Pointers.CPUIDFunction = PMF.GetConvertedPointer();
-    }
-
-    Pointers.SyscallHandlerObj = reinterpret_cast<uint64_t>(CTX->SyscallHandler);
-    Pointers.SyscallHandlerFunc = reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall);
-
-    // Fill in the fallback handlers
-    InterpreterOps::FillFallbackIndexPointers(Pointers.FallbackHandlerPointers);
-
-    // Thread Specific
-    Pointers.SignalHandlerRefCountPointer = reinterpret_cast<uint64_t>(&Dispatcher->SignalHandlerRefCounter);
-  }
-
-  {
-    DispatcherConfig config;
-    config.ExitFunctionLink = reinterpret_cast<uintptr_t>(&ExitFunctionLink);
-    config.ExitFunctionLinkThis = reinterpret_cast<uintptr_t>(this);
-    config.StaticRegisterAssignment = ctx->Config.StaticRegisterAllocation;
-
-    Dispatcher = std::make_unique<Arm64Dispatcher>(CTX, ThreadState, config);
-    DispatchPtr = Dispatcher->DispatchPtr;
-    CallbackPtr = Dispatcher->CallbackPtr;
-  }
-
-  // Can't allocate a code buffer until after dispatcher is created
-  InitialCodeBuffer = AllocateNewCodeBuffer(Arm64JITCore::INITIAL_CODE_SIZE);
-  *GetBuffer() = vixl::CodeBuffer(InitialCodeBuffer.Ptr, InitialCodeBuffer.Size);
-  SetAllowAssembler(true);
-
-  CurrentCodeBuffer = &InitialCodeBuffer;
+Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
+  : CPUBackend(Thread, INITIAL_CODE_SIZE, MAX_CODE_SIZE)
+  , Arm64Emitter(ctx, 0)
+  , CTX {ctx} {
 
   RAPass = Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA");
 
@@ -447,95 +465,76 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::Intern
   RegisterVectorHandlers();
   RegisterEncryptionHandlers();
 
-  if (!CompileThread) {
-    ThreadSharedData.SignalHandlerRefCounterPtr = &Dispatcher->SignalHandlerRefCounter;
-    ThreadSharedData.SignalReturnInstruction = Dispatcher->SignalHandlerReturnAddress;
-    ThreadSharedData.UnimplementedInstructionAddress = Dispatcher->UnimplementedInstructionAddress;
+  {
+    // Set up pointers that the JIT needs to load
 
-    ThreadSharedData.Dispatcher = Dispatcher.get();
+    // Common
+    auto &Common = ThreadState->CurrentFrame->Pointers.Common;
+    
+    Common.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
+    Common.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
+    Common.RemoveThreadCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveThreadCodeEntryFromJit);
+    Common.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
 
-    // This will register the host signal handler per thread, which is fine
-    CTX->SignalDelegation->RegisterHostSignalHandler(SIGILL, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      Arm64JITCore *Core = reinterpret_cast<Arm64JITCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleSIGILL(Signal, info, ucontext);
-    }, true);
-
-    CTX->SignalDelegation->RegisterHostSignalHandler(SIGBUS, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      Arm64JITCore *Core = reinterpret_cast<Arm64JITCore*>(Thread->CPUBackend.get());
-
-      if (!Core->Dispatcher->IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext))) {
-        // Wasn't a sigbus in JIT code
-        return false;
-      }
-
-      return FEXCore::ArchHelpers::Arm64::HandleSIGBUS(Core->CTX->Config.ParanoidTSO(), Signal, info, ucontext);
-    }, true);
-
-    CTX->SignalDelegation->RegisterHostSignalHandler(SignalDelegator::SIGNAL_FOR_PAUSE, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      Arm64JITCore *Core = reinterpret_cast<Arm64JITCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleSignalPause(Signal, info, ucontext);
-    }, true);
-
-    auto GuestSignalHandler = [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) -> bool {
-      Arm64JITCore *Core = reinterpret_cast<Arm64JITCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleGuestSignal(Signal, info, ucontext, GuestAction, GuestStack);
-    };
-
-    for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
-      CTX->SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
+    {
+      FEXCore::Utils::MemberFunctionToPointerCast PMF(&FEXCore::CPUIDEmu::RunFunction);
+      Common.CPUIDFunction = PMF.GetConvertedPointer();
     }
+
+    Common.SyscallHandlerObj = reinterpret_cast<uint64_t>(CTX->SyscallHandler);
+    Common.SyscallHandlerFunc = reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall);
+    Common.ExitFunctionLink = reinterpret_cast<uintptr_t>(&Arm64JITCore_ExitFunctionLink);
+
+
+    // Fill in the fallback handlers
+    InterpreterOps::FillFallbackIndexPointers(Common.FallbackHandlerPointers);
+
+    // Platform Specific
+    auto &AArch64 = ThreadState->CurrentFrame->Pointers.AArch64;
+    
+    AArch64.LUDIV = reinterpret_cast<uint64_t>(LUDIV);
+    AArch64.LDIV = reinterpret_cast<uint64_t>(LDIV);
+    AArch64.LUREM = reinterpret_cast<uint64_t>(LUREM);
+    AArch64.LREM = reinterpret_cast<uint64_t>(LREM);
   }
+
+  // Must be done after Dispatcher init
+  SetAllowAssembler(true);
+  ClearCache();
+}
+
+void Arm64JITCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
+  CTX->SignalDelegation->RegisterHostSignalHandler(SIGILL, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
+    return Thread->CTX->Dispatcher->HandleSIGILL(Thread, Signal, info, ucontext);
+  }, true);
+
+  CTX->SignalDelegation->RegisterHostSignalHandler(SIGBUS, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
+    if (!Thread->CPUBackend->IsAddressInCodeBuffer(ArchHelpers::Context::GetPc(ucontext))) {
+      // Wasn't a sigbus in JIT code
+      return false;
+    }
+
+    return FEXCore::ArchHelpers::Arm64::HandleSIGBUS(Thread->CTX->Config.ParanoidTSO(), Signal, info, ucontext);
+  }, true);
+}
+
+void Arm64JITCore::EmitDetectionString() {
+  const char JITString[] = "FEXJIT::Arm64JITCore::";
+  auto Buffer = GetBuffer();
+  Buffer->EmitString(JITString);
+  Buffer->Align();
 }
 
 void Arm64JITCore::ClearCache() {
   // Get the backing code buffer
-  auto Buffer = GetBuffer();
-  if (*ThreadSharedData.SignalHandlerRefCounterPtr == 0) {
-    if (!CodeBuffers.empty()) {
-      // If we have more than one code buffer we are tracking then walk them and delete
-      // This is a cleanup step
-      for (auto CodeBuffer : CodeBuffers) {
-        FreeCodeBuffer(CodeBuffer);
-      }
-      CodeBuffers.clear();
-
-      // Set the current code buffer to the initial
-      *Buffer = vixl::CodeBuffer(InitialCodeBuffer.Ptr, InitialCodeBuffer.Size);
-      CurrentCodeBuffer = &InitialCodeBuffer;
-    }
-
-    if (CurrentCodeBuffer->Size == MAX_CODE_SIZE) {
-      // Rewind to the start of the code cache start
-      Buffer->Reset();
-    }
-    else {
-      FreeCodeBuffer(InitialCodeBuffer);
-
-      // Resize the code buffer and reallocate our code size
-      InitialCodeBuffer.Size *= 1.5;
-      InitialCodeBuffer.Size = std::min(InitialCodeBuffer.Size, MAX_CODE_SIZE);
-
-      InitialCodeBuffer = AllocateNewCodeBuffer(InitialCodeBuffer.Size);
-      *Buffer = vixl::CodeBuffer(InitialCodeBuffer.Ptr, InitialCodeBuffer.Size);
-    }
-  }
-  else {
-    // We have signal handlers that have generated code
-    // This means that we can not safely clear the code at this point in time
-    // Allocate some new code buffers that we can switch over to instead
-    auto NewCodeBuffer = Arm64JITCore::AllocateNewCodeBuffer(Arm64JITCore::INITIAL_CODE_SIZE);
-    EmplaceNewCodeBuffer(NewCodeBuffer);
-    *Buffer = vixl::CodeBuffer(NewCodeBuffer.Ptr, NewCodeBuffer.Size);
-  }
+  
+  auto CodeBuffer = GetEmptyCodeBuffer();
+  *GetBuffer() = vixl::CodeBuffer(CodeBuffer->Ptr, CodeBuffer->Size);
+  EmitDetectionString();
 }
 
 Arm64JITCore::~Arm64JITCore() {
-  for (auto CodeBuffer : CodeBuffers) {
-    FreeCodeBuffer(CodeBuffer);
-  }
-  CodeBuffers.clear();
 
-  FreeCodeBuffer(InitialCodeBuffer);
 }
 
 IR::PhysicalRegister Arm64JITCore::GetPhys(IR::NodeID Node) const {
@@ -666,13 +665,14 @@ bool Arm64JITCore::IsGPR(IR::NodeID Node) const {
   return Class == IR::GPRClass || Class == IR::GPRFixedClass;
 }
 
-void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData) {
+void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) {
   using namespace aarch64;
   JumpTargets.clear();
   uint32_t SSACount = IR->GetSSACount();
 
   this->Entry = Entry;
   this->RAData = RAData;
+  this->DebugData = DebugData;
 
   #ifndef NDEBUG
   LoadConstant(x0, Entry);
@@ -681,9 +681,9 @@ void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IR
   this->IR = IR;
 
   // Fairly excessive buffer range to make sure we don't overflow
-  uint32_t BufferRange = SSACount * 16;
+  uint32_t BufferRange = SSACount * 16 + GDBEnabled * Dispatcher::MaxGDBPauseCheckSize;
   if ((GetCursorOffset() + BufferRange) > CurrentCodeBuffer->Size) {
-    ThreadState->CTX->ClearCodeCache(ThreadState, false);
+    CTX->ClearCodeCache(ThreadState);
   }
 
   // AAPCS64
@@ -706,31 +706,11 @@ void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IR
   // X1-X3 = Temp
   // X4-r18 = RA
 
-  auto GuestEntry = GetCursorAddress<uint64_t>();
+  GuestEntry = GetCursorAddress<uint8_t *>();
 
- if (CTX->GetGdbServerStatus()) {
-    aarch64::Label RunBlock;
-
-    // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
-    // This happens when single stepping
-
-    static_assert(sizeof(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-    ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Thread))); // Get thread
-    ldr(x0, MemOperand(x0, offsetof(FEXCore::Core::InternalThreadState, CTX))); // Get Context
-    ldr(w0, MemOperand(x0, offsetof(FEXCore::Context::Context, Config.RunningMode)));
-
-    // If the value == 0 then we don't need to stop
-    cbz(w0, &RunBlock);
-    {
-      // Make sure RIP is syncronized to the context
-      LoadConstant(x0, Entry);
-      str(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
-
-      // Stop the thread
-      ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.ThreadPauseHandlerSpillSRA)));
-      br(x0);
-    }
-    bind(&RunBlock);
+  if (GDBEnabled) {
+    auto GDBSize = CTX->Dispatcher->GenerateGDBPauseCheck(GuestEntry, Entry);
+    GetBuffer()->CursorForward(GDBSize);
   }
 
   //LOGMAN_THROW_A_FMT(RAData->HasFullRA(), "Arm64 JIT only works with RA");
@@ -755,6 +735,7 @@ void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IR
     LOGMAN_THROW_A_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");
 #endif
 
+    auto BlockStartHostCode = GetCursorAddress<uint8_t *>();
     {
       const auto Node = IR->GetID(BlockNode);
       const auto IsTarget = JumpTargets.try_emplace(Node).first;
@@ -769,10 +750,6 @@ void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IR
       bind(&IsTarget->second);
     }
 
-    if (DebugData) {
-      DebugData->Subblocks.push_back({GetCursorAddress<uintptr_t>(), 0, IR->GetID(BlockNode)});
-    }
-
     for (auto [CodeNode, IROp] : IR->GetCode(BlockNode)) {
       const auto ID = IR->GetID(CodeNode);
 
@@ -782,7 +759,10 @@ void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IR
     }
 
     if (DebugData) {
-      DebugData->Subblocks.back().HostCodeSize = GetCursorAddress<uintptr_t>() - DebugData->Subblocks.back().HostCodeStart;
+      DebugData->Subblocks.push_back({
+        static_cast<uint32_t>(BlockStartHostCode - GuestEntry),
+        static_cast<uint32_t>(GetCursorAddress<uint8_t *>() - BlockStartHostCode)
+      });
     }
   }
 
@@ -795,68 +775,44 @@ void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IR
 
   FinalizeCode();
 
-  auto CodeEnd = GetCursorAddress<uint64_t>();
-  CPU.EnsureIAndDCacheCoherency(reinterpret_cast<void*>(GuestEntry), CodeEnd - reinterpret_cast<uint64_t>(GuestEntry));
+  auto CodeEnd = GetCursorAddress<uint8_t *>();
+  CPU.EnsureIAndDCacheCoherency(GuestEntry, CodeEnd - GuestEntry);
 
   if (DebugData) {
-    DebugData->HostCodeSize = reinterpret_cast<uintptr_t>(CodeEnd) - reinterpret_cast<uintptr_t>(GuestEntry);
+    DebugData->HostCodeSize = CodeEnd - GuestEntry;
+    DebugData->Relocations = &Relocations;
   }
 
   this->IR = nullptr;
 
-  return reinterpret_cast<void*>(GuestEntry);
+  return GuestEntry;
 }
 
-uint64_t Arm64JITCore::ExitFunctionLink(Arm64JITCore *core, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
-  auto Thread = Frame->Thread;
-  auto GuestRip = record[1];
+void Arm64JITCore::ResetStack() {
+  if (SpillSlots == 0)
+    return;
 
-  auto HostCode = Thread->LookupCache->FindBlock(GuestRip);
-
-  if (!HostCode) {
-    //fmt::print("ExitFunctionLink: Aborting, {:X} not in cache\n", GuestRip);
-    Frame->State.rip = GuestRip;
-    return core->ThreadSharedData.Dispatcher->AbsoluteLoopTopAddress;
-  }
-
-  uintptr_t branch = (uintptr_t)(record) - 8;
-  auto LinkerAddress = core->ThreadSharedData.Dispatcher->ExitFunctionLinkerAddress;
-
-  auto offset = HostCode/4 - branch/4;
-  if (IsInt26(offset)) {
-    // optimal case - can branch directly
-    // patch the code
-    vixl::aarch64::Assembler emit((uint8_t*)(branch), 24);
-    vixl::CodeBufferCheckScope scope(&emit, 24, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
-    emit.b(offset);
-    emit.FinalizeCode();
-    vixl::aarch64::CPU::EnsureIAndDCacheCoherency((void*)branch, 24);
-
-    // Add de-linking handler
-    Thread->LookupCache->AddBlockLink(GuestRip, (uintptr_t)record, [branch, LinkerAddress]{
-      vixl::aarch64::Assembler emit((uint8_t*)(branch), 24);
-      vixl::CodeBufferCheckScope scope(&emit, 24, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
-      Literal l_BranchHost{LinkerAddress};
-      emit.ldr(x0, &l_BranchHost);
-      emit.blr(x0);
-      emit.place(&l_BranchHost);
-      emit.FinalizeCode();
-      vixl::aarch64::CPU::EnsureIAndDCacheCoherency((void*)branch, 24);
-    });
+  if (IsImmAddSub(SpillSlots * 16)) {
+    add(sp, sp, SpillSlots * 16);
   } else {
-    // fallback case - do a soft-er link by patching the pointer
-    record[0] = HostCode;
-
-    // Add de-linking handler
-    Thread->LookupCache->AddBlockLink(GuestRip, (uintptr_t)record, [record, LinkerAddress]{
-      record[0] = LinkerAddress;
-    });
+   // Too big to fit in a 12bit immediate
+   LoadConstant(x0, SpillSlots * 16);
+   add(sp, sp, x0);
   }
-
-  return HostCode;
 }
 
-std::unique_ptr<CPUBackend> CreateArm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread) {
-  return std::make_unique<Arm64JITCore>(ctx, Thread, CompileThread);
+std::unique_ptr<CPUBackend> CreateArm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
+  return std::make_unique<Arm64JITCore>(ctx, Thread);
 }
+
+void InitializeArm64JITSignalHandlers(FEXCore::Context::Context *CTX) {
+  Arm64JITCore::InitializeSignalHandlers(CTX);
+}
+
+CPUBackendFeatures GetArm64JITBackendFeatures() {
+  return CPUBackendFeatures {
+    .SupportsStaticRegisterAllocation = true
+  };
+}
+
 }

External/FEXCore/Source/Interface/Core/JIT/Arm64/JITClass.h

@@ -6,17 +6,23 @@ $end_info$
 
 #pragma once
 
+#include <FEXCore/IR/RegisterAllocationData.h>
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 
-#include "aarch64/assembler-aarch64.h"
-#include "aarch64/disasm-aarch64.h"
-#include "aarch64/assembler-aarch64.h"
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/disasm-aarch64.h>
 
 #include <FEXCore/Core/CPUBackend.h>
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 
+#include <array>
+#include <cstdint>
+#include <map>
+#include <utility>
+#include <vector>
+
 #define STATE x28
 #define TMP1 x0
 #define TMP2 x1
@@ -37,14 +43,8 @@ using namespace vixl::aarch64;
 
 class Arm64JITCore final : public CPUBackend, public Arm64Emitter  {
 public:
-  struct CodeBuffer {
-    uint8_t *Ptr;
-    size_t Size;
-  };
-
   explicit Arm64JITCore(FEXCore::Context::Context *ctx,
-                        FEXCore::Core::InternalThreadState *Thread,
-                        bool CompileThread);
+                        FEXCore::Core::InternalThreadState *Thread);
   ~Arm64JITCore() override;
 
   [[nodiscard]] std::string GetName() override { return "JIT"; }
@@ -52,7 +52,7 @@ public:
   [[nodiscard]] void *CompileCode(uint64_t Entry,
                                   FEXCore::IR::IRListView const *IR,
                                   FEXCore::Core::DebugData *DebugData,
-                                  FEXCore::IR::RegisterAllocationData *RAData) override;
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;
 
   [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }
 
@@ -60,21 +60,15 @@ public:
 
   void ClearCache() override;
 
-  static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
-  [[nodiscard]] CodeBuffer AllocateNewCodeBuffer(size_t Size);
+  static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
 
-  void CopyNecessaryDataForCompileThread(CPUBackend *Original) override;
-  bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true, bool IncludeCompileService = true) const override {
-    return Dispatcher->IsAddressInJITCode(Address, IncludeDispatcher, IncludeCompileService);
-  }
+  void ClearRelocations() override { Relocations.clear(); }
 
 private:
   FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
 
-  std::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
   Label *PendingTargetLabel;
   FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *ThreadState;
   FEXCore::IR::IRListView const *IR;
   uint64_t Entry;
 
@@ -145,46 +139,77 @@ private:
   vixl::aarch64::Decoder Decoder;
 #endif
 
-  void EmplaceNewCodeBuffer(CodeBuffer Buffer) {
-    CurrentCodeBuffer = &CodeBuffers.emplace_back(Buffer);
-  }
-
-  void FreeCodeBuffer(CodeBuffer Buffer);
-
-  // This is the initial code buffer that we will fall back to
-  // In a program without signals and code clearing, we will typically
-  // only have this code buffer
-  CodeBuffer InitialCodeBuffer{};
-  // This is the array of /additional/ code buffers that we may need to allocate
-  // Allocation only occurs when we've hit signals and need to clear code cache
-  // For code safety we can't delete code buffers until outside of all signals
-  std::vector<CodeBuffer> CodeBuffers{};
-
-  // This is the current code buffer that we are tracking
-  CodeBuffer *CurrentCodeBuffer{};
-
-  // We don't want to mvoe above 128MB atm because that means we will have to encode longer jumps
-  static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 128;
-  static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096 * 2;
-
 #if DEBUG
   vixl::aarch64::Disassembler Disasm;
 #endif
 
-  static uint64_t ExitFunctionLink(Arm64JITCore *core, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record);
-
-  struct CompilerSharedData {
-    uint64_t SignalReturnInstruction{};
-    uint64_t UnimplementedInstructionAddress{};
-    uint64_t OverflowExceptionInstructionAddress{};
-
-    uint32_t *SignalHandlerRefCounterPtr{};
-    FEXCore::CPU::Dispatcher *Dispatcher{};
-  };
-
-  CompilerSharedData ThreadSharedData;
+  // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
+  void EmitDetectionString();
   IR::RegisterAllocationPass *RAPass;
   IR::RegisterAllocationData *RAData;
+  FEXCore::Core::DebugData *DebugData;
+
+  void ResetStack();
+  /**
+   * @name Relocations
+   * @{ */
+
+    uint64_t GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+
+    /**
+     * @brief A literal pair relocation object for named symbol literals
+     */
+    struct NamedSymbolLiteralPair {
+      Literal<uint64_t> Lit;
+      Relocation MoveABI{};
+    };
+
+    /**
+     * @brief Inserts a thunk relocation
+     *
+     * @param Reg - The GPR to move the thunk handler in to
+     * @param Sum - The hash of the thunk
+     */
+    void InsertNamedThunkRelocation(vixl::aarch64::Register Reg, const IR::SHA256Sum &Sum);
+
+    /**
+     * @brief Inserts a guest GPR move relocation
+     *
+     * @param Reg - The GPR to move the guest RIP in to
+     * @param Constant - The guest RIP that will be relocated
+     */
+    void InsertGuestRIPMove(vixl::aarch64::Register Reg, uint64_t Constant);
+
+    /**
+     * @brief Inserts a named symbol as a literal in memory
+     *
+     * Need to use `PlaceNamedSymbolLiteral` with the return value to place the literal in the desired location
+     *
+     * @param Op The named symbol to place
+     *
+     * @return A temporary `NamedSymbolLiteralPair`
+     */
+    NamedSymbolLiteralPair InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+
+    /**
+     * @brief Place the named symbol literal relocation in memory
+     *
+     * @param Lit - Which literal to place
+     */
+    void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit);
+
+    std::vector<FEXCore::CPU::Relocation> Relocations;
+
+    ///< Relocation code loading
+    bool ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations);
+
+  /**  @} */
+
+  uint32_t SpillSlots{};
+  /**
+  * @brief Current guest RIP entrypoint
+  */
+  uint8_t *GuestEntry{};
 
   using OpHandler = void (Arm64JITCore::*)(IR::IROp_Header *IROp, IR::NodeID Node);
   std::array<OpHandler, IR::IROps::OP_LAST + 1> OpHandlers {};
@@ -275,8 +300,6 @@ private:
   DEF_OP(AtomicFetchNeg);
 
   ///< Branch ops
-  DEF_OP(GuestCallDirect);
-  DEF_OP(GuestCallIndirect);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -286,7 +309,7 @@ private:
   DEF_OP(InlineSyscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveCodeEntry);
+  DEF_OP(RemoveThreadCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops
@@ -326,7 +349,7 @@ private:
   DEF_OP(CacheLineZero);
 
   ///< Misc ops
-  DEF_OP(EndBlock);
+  DEF_OP(GuestOpcode);
   DEF_OP(Fence);
   DEF_OP(Break);
   DEF_OP(Phi);
@@ -336,6 +359,7 @@ private:
   DEF_OP(SetRoundingMode);
   DEF_OP(ProcessorID);
   DEF_OP(RDRAND);
+  DEF_OP(Yield);
 
   ///< Move ops
   DEF_OP(ExtractElementPair);
@@ -442,9 +466,8 @@ private:
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
   DEF_OP(CRC32);
+  DEF_OP(PCLMUL);
 #undef DEF_OP
 };
 
-
-}
-
+} // namespace FEXCore::CPU

External/FEXCore/Source/Interface/Core/JIT/Arm64/MemoryOps.cpp

@@ -4,6 +4,8 @@ tags: backend|arm64
 $end_info$
 */
 
+#include "Interface/Context/Context.h"
+#include "Interface/Core/CPUID.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 #include <FEXCore/Utils/CompilerDefs.h>
 
@@ -58,26 +60,26 @@ DEF_OP(LoadContext) {
 
 DEF_OP(StoreContext) {
   auto Op = IROp->C<IR::IROp_StoreContext>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {
     case 1:
-      strb(GetReg<RA_32>(Op->Header.Args[0].ID()), MemOperand(STATE, Op->Offset));
+      strb(GetReg<RA_32>(Op->Value.ID()), MemOperand(STATE, Op->Offset));
     break;
     case 2:
-      strh(GetReg<RA_32>(Op->Header.Args[0].ID()), MemOperand(STATE, Op->Offset));
+      strh(GetReg<RA_32>(Op->Value.ID()), MemOperand(STATE, Op->Offset));
     break;
     case 4:
-      str(GetReg<RA_32>(Op->Header.Args[0].ID()), MemOperand(STATE, Op->Offset));
+      str(GetReg<RA_32>(Op->Value.ID()), MemOperand(STATE, Op->Offset));
     break;
     case 8:
-      str(GetReg<RA_64>(Op->Header.Args[0].ID()), MemOperand(STATE, Op->Offset));
+      str(GetReg<RA_64>(Op->Value.ID()), MemOperand(STATE, Op->Offset));
     break;
     default:  LOGMAN_MSG_A_FMT("Unhandled StoreContext size: {}", OpSize);
     }
   }
   else {
-    auto Src =  GetSrc(Op->Header.Args[0].ID());
+    auto Src =  GetSrc(Op->Value.ID());
     switch (OpSize) {
     case 1:
       str(Src.B(), MemOperand(STATE, Op->Offset));
@@ -104,7 +106,7 @@ DEF_OP(LoadRegister) {
   auto Op = IROp->C<IR::IROp_LoadRegister>();
 
   if (Op->Class == IR::GPRClass) {
-    auto regId = (Op->Offset - offsetof(FEXCore::Core::CpuStateFrame, State.gregs[0])) / 8;
+    auto regId = (Op->Offset - offsetof(Core::CpuStateFrame, State.gregs[0])) / Core::CPUState::GPR_REG_SIZE;
     auto regOffs = Op->Offset & 7;
 
     LOGMAN_THROW_A_FMT(regId < SRA64.size(), "out of range regId");
@@ -135,7 +137,7 @@ DEF_OP(LoadRegister) {
         break;
     }
   } else if (Op->Class == IR::FPRClass) {
-    auto regId = (Op->Offset - offsetof(FEXCore::Core::CpuStateFrame, State.xmm[0][0])) / 16;
+    auto regId = (Op->Offset - offsetof(Core::CpuStateFrame, State.xmm[0][0])) / Core::CPUState::XMM_REG_SIZE;
     auto regOffs = Op->Offset & 15;
 
     LOGMAN_THROW_A_FMT(regId < SRAFPR.size(), "out of range regId");
@@ -189,7 +191,7 @@ DEF_OP(StoreRegister) {
   auto Op = IROp->C<IR::IROp_StoreRegister>();
 
   if (Op->Class == IR::GPRClass) {
-    auto regId = Op->Offset / 8 - 1;
+    auto regId = (Op->Offset / Core::CPUState::GPR_REG_SIZE) - 1;
     auto regOffs = Op->Offset & 7;
 
     LOGMAN_THROW_A_FMT(regId < SRA64.size(), "out of range regId");
@@ -219,7 +221,7 @@ DEF_OP(StoreRegister) {
         break;
     }
   } else if (Op->Class == IR::FPRClass) {
-    auto regId = (Op->Offset - offsetof(FEXCore::Core::CpuStateFrame, State.xmm[0][0])) / 16;
+    auto regId = (Op->Offset - offsetof(Core::CpuStateFrame, State.xmm[0][0])) / Core::CPUState::XMM_REG_SIZE;
     auto regOffs = Op->Offset & 15;
 
     LOGMAN_THROW_A_FMT(regId < SRAFPR.size(), "regId out of range");
@@ -261,8 +263,8 @@ DEF_OP(StoreRegister) {
 
 DEF_OP(LoadContextIndexed) {
   auto Op = IROp->C<IR::IROp_LoadContextIndexed>();
-  size_t size = IROp->Size;
-  auto index = GetReg<RA_64>(Op->Header.Args[0].ID());
+  const size_t size = IROp->Size;
+  auto index = GetReg<RA_64>(Op->Index.ID());
 
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (Op->Stride) {
@@ -349,11 +351,11 @@ DEF_OP(LoadContextIndexed) {
 
 DEF_OP(StoreContextIndexed) {
   auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
-  size_t size = IROp->Size;
-  auto index = GetReg<RA_64>(Op->Header.Args[1].ID());
+  const size_t size = IROp->Size;
+  auto index = GetReg<RA_64>(Op->Index.ID());
 
   if (Op->Class == FEXCore::IR::GPRClass) {
-    auto value = GetReg<RA_64>(Op->Header.Args[0].ID());
+    auto value = GetReg<RA_64>(Op->Value.ID());
 
     switch (Op->Stride) {
     case 1:
@@ -392,7 +394,7 @@ DEF_OP(StoreContextIndexed) {
     }
   }
   else {
-    auto value = GetSrc(Op->Header.Args[0].ID());
+    auto value = GetSrc(Op->Value.ID());
 
     switch (Op->Stride) {
     case 1:
@@ -441,25 +443,25 @@ DEF_OP(StoreContextIndexed) {
 
 DEF_OP(SpillRegister) {
   auto Op = IROp->C<IR::IROp_SpillRegister>();
-  uint8_t OpSize = IROp->Size;
-  uint32_t SlotOffset = Op->Slot * 16;
+  const uint8_t OpSize = IROp->Size;
+  const uint32_t SlotOffset = Op->Slot * 16;
 
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {
     case 1: {
-      strb(GetReg<RA_64>(Op->Header.Args[0].ID()), MemOperand(sp, SlotOffset));
+      strb(GetReg<RA_64>(Op->Value.ID()), MemOperand(sp, SlotOffset));
       break;
     }
     case 2: {
-      strh(GetReg<RA_64>(Op->Header.Args[0].ID()), MemOperand(sp, SlotOffset));
+      strh(GetReg<RA_64>(Op->Value.ID()), MemOperand(sp, SlotOffset));
       break;
     }
     case 4: {
-      str(GetReg<RA_32>(Op->Header.Args[0].ID()), MemOperand(sp, SlotOffset));
+      str(GetReg<RA_32>(Op->Value.ID()), MemOperand(sp, SlotOffset));
       break;
     }
     case 8: {
-      str(GetReg<RA_64>(Op->Header.Args[0].ID()), MemOperand(sp, SlotOffset));
+      str(GetReg<RA_64>(Op->Value.ID()), MemOperand(sp, SlotOffset));
       break;
     }
     default:  LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize);
@@ -467,15 +469,15 @@ DEF_OP(SpillRegister) {
   } else if (Op->Class == FEXCore::IR::FPRClass) {
     switch (OpSize) {
     case 4: {
-      str(GetSrc(Op->Header.Args[0].ID()).S(), MemOperand(sp, SlotOffset));
+      str(GetSrc(Op->Value.ID()).S(), MemOperand(sp, SlotOffset));
       break;
     }
     case 8: {
-      str(GetSrc(Op->Header.Args[0].ID()).D(), MemOperand(sp, SlotOffset));
+      str(GetSrc(Op->Value.ID()).D(), MemOperand(sp, SlotOffset));
       break;
     }
     case 16: {
-      str(GetSrc(Op->Header.Args[0].ID()), MemOperand(sp, SlotOffset));
+      str(GetSrc(Op->Value.ID()), MemOperand(sp, SlotOffset));
       break;
     }
     default:  LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize);
@@ -539,7 +541,7 @@ DEF_OP(LoadFlag) {
 
 DEF_OP(StoreFlag) {
   auto Op = IROp->C<IR::IROp_StoreFlag>();
-  strb(GetReg<RA_64>(Op->Header.Args[0].ID()), MemOperand(STATE, offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag));
+  strb(GetReg<RA_64>(Op->Value.ID()), MemOperand(STATE, offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag));
 }
 
 MemOperand Arm64JITCore::GenerateMemOperand(uint8_t AccessSize, aarch64::Register Base, IR::OrderedNodeWrapper Offset, IR::MemOffsetType OffsetType, uint8_t OffsetScale) {
@@ -570,7 +572,7 @@ MemOperand Arm64JITCore::GenerateMemOperand(uint8_t AccessSize, aarch64::Registe
 DEF_OP(LoadMem) {
   auto Op = IROp->C<IR::IROp_LoadMem>();
 
-  auto MemReg = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemReg = GetReg<RA_64>(Op->Addr.ID());
   auto MemSrc = GenerateMemOperand(IROp->Size, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
 
   if (Op->Class == FEXCore::IR::GPRClass) {
@@ -617,13 +619,43 @@ DEF_OP(LoadMem) {
 DEF_OP(LoadMemTSO) {
   auto Op = IROp->C<IR::IROp_LoadMemTSO>();
 
-  auto MemSrc = MemOperand(GetReg<RA_64>(Op->Header.Args[0].ID()));
+  auto MemReg = GetReg<RA_64>(Op->Addr.ID());
+  auto MemSrc = GenerateMemOperand(IROp->Size, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
 
-  if (!Op->Offset.IsInvalid()) {
-    LOGMAN_MSG_A_FMT("LoadMemTSO: No offset allowed");
+  if (CTX->HostFeatures.SupportsTSOImm9) {
+    // RCPC2 means that the offset must be an inline constant
+    LOGMAN_THROW_A_FMT(MemSrc.IsRegisterOffset() == false, "RCPC2 doesn't support register offset. Only Immediate offset");
+  }
+  else {
+    LOGMAN_THROW_A_FMT(Op->Offset.IsInvalid(), "LoadMemTSO: No offset allowed");
   }
 
-  if (CTX->HostFeatures.SupportsRCPC && Op->Class == FEXCore::IR::GPRClass) {
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
+    if (IROp->Size == 1) {
+      // 8bit load is always aligned to natural alignment
+      auto Dst = GetReg<RA_64>(Node);
+      ldapurb(Dst, MemSrc);
+    }
+    else {
+      // Aligned
+      nop();
+      auto Dst = GetReg<RA_64>(Node);
+      switch (IROp->Size) {
+        case 2:
+          ldapurh(Dst, MemSrc);
+          break;
+        case 4:
+          ldapur(Dst.W(), MemSrc);
+          break;
+        case 8:
+          ldapur(Dst, MemSrc);
+          break;
+        default:  LOGMAN_MSG_A_FMT("Unhandled LoadMemTSO size: {}", IROp->Size);
+      }
+      nop();
+    }
+  }
+  else if (CTX->HostFeatures.SupportsRCPC && Op->Class == FEXCore::IR::GPRClass) {
     if (IROp->Size == 1) {
       // 8bit load is always aligned to natural alignment
       auto Dst = GetReg<RA_64>(Node);
@@ -744,13 +776,41 @@ DEF_OP(StoreMem) {
 
 DEF_OP(StoreMemTSO) {
   auto Op = IROp->C<IR::IROp_StoreMemTSO>();
-  auto MemSrc = MemOperand(GetReg<RA_64>(Op->Addr.ID()));
 
-  if (!Op->Offset.IsInvalid()) {
-    LOGMAN_MSG_A_FMT("StoreMemTSO: No offset allowed");
+  auto MemReg = GetReg<RA_64>(Op->Addr.ID());
+  auto MemSrc = GenerateMemOperand(IROp->Size, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
+
+  if (CTX->HostFeatures.SupportsTSOImm9) {
+    // RCPC2 means that the offset must be an inline constant
+    LOGMAN_THROW_A_FMT(MemSrc.IsRegisterOffset() == false, "RCPC2 doesn't support register offset. Only Immediate offset");
+  }
+  else {
+    LOGMAN_THROW_A_FMT(Op->Offset.IsInvalid(), "StoreMemTSO: No offset allowed");
   }
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
+    if (IROp->Size == 1) {
+      // 8bit load is always aligned to natural alignment
+      stlurb(GetReg<RA_64>(Op->Value.ID()), MemSrc);
+    }
+    else {
+      nop();
+      switch (IROp->Size) {
+        case 2:
+          stlurh(GetReg<RA_64>(Op->Value.ID()), MemSrc);
+          break;
+        case 4:
+          stlur(GetReg<RA_32>(Op->Value.ID()), MemSrc);
+          break;
+        case 8:
+          stlur(GetReg<RA_64>(Op->Value.ID()), MemSrc);
+          break;
+        default:  LOGMAN_MSG_A_FMT("Unhandled StoreMemTSO size: {}", IROp->Size);
+      }
+      nop();
+    }
+  }
+  else if (Op->Class == FEXCore::IR::GPRClass) {
     if (IROp->Size == 1) {
       // 8bit load is always aligned to natural alignment
       stlrb(GetReg<RA_64>(Op->Value.ID()), MemSrc);
@@ -934,7 +994,7 @@ DEF_OP(VStoreMemElement) {
 DEF_OP(CacheLineClear) {
   auto Op = IROp->C<IR::IROp_CacheLineClear>();
 
-  auto MemReg = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemReg = GetReg<RA_64>(Op->Addr.ID());
 
   // Clear dcache only
   // icache doesn't matter here since the guest application shouldn't be calling clflush on JIT code.
@@ -949,7 +1009,7 @@ DEF_OP(CacheLineClear) {
 DEF_OP(CacheLineZero) {
   auto Op = IROp->C<IR::IROp_CacheLineZero>();
 
-  auto MemReg = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemReg = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsCLZERO) {
     // We can use this instruction directly

External/FEXCore/Source/Interface/Core/JIT/Arm64/MiscOps.cpp

@@ -4,13 +4,21 @@ tags: backend|arm64
 $end_info$
 */
 
+#include <syscall.h>
 #include "Interface/Core/JIT/Arm64/JITClass.h"
+#include "FEXCore/Debug/InternalThreadState.h"
 
 namespace FEXCore::CPU {
 using namespace vixl;
 using namespace vixl::aarch64;
 #define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 
+DEF_OP(GuestOpcode) {
+  auto Op = IROp->C<IR::IROp_GuestOpcode>();
+  // metadata
+  DebugData->GuestOpcodes.push_back({Op->GuestEntryOffset, GetCursorAddress<uint8_t*>() - GuestEntry});
+}
+
 DEF_OP(Fence) {
   auto Op = IROp->C<IR::IROp_Fence>();
   switch (Op->Fence) {
@@ -36,7 +44,7 @@ DEF_OP(Break) {
       break;
     case FEXCore::IR::Break_Overflow: // overflow
       ResetStack();
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.OverflowExceptionHandler)));
+      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.OverflowExceptionHandler)));
       br(TMP1);
       break;
     case FEXCore::IR::Break_Halt: { // HLT
@@ -46,13 +54,13 @@ DEF_OP(Break) {
       add(sp, TMP1, 0);
 
       // Now we need to jump to the thread stop handler
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.ThreadStopHandlerSpillSRA)));
+      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadStopHandlerSpillSRA)));
       br(TMP1);
       break;
     }
     case FEXCore::IR::Break_Interrupt3: { // INT3
       ResetStack();
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.ThreadPauseHandlerSpillSRA)));
+      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA)));
       br(TMP1);
       break;
     }
@@ -60,7 +68,7 @@ DEF_OP(Break) {
     {
       ResetStack();
 
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.UnimplementedInstructionHandler)));
+      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.UnimplementedInstructionHandler)));
       br(TMP1);
 
       break;
@@ -97,7 +105,7 @@ DEF_OP(GetRoundingMode) {
 
 DEF_OP(SetRoundingMode) {
   auto Op = IROp->C<IR::IROp_SetRoundingMode>();
-  auto Src = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto Src = GetReg<RA_64>(Op->RoundMode.ID());
 
   // Setup the rounding flags correctly
   and_(TMP1, Src, 0b11);
@@ -132,15 +140,15 @@ DEF_OP(Print) {
 
   PushDynamicRegsAndLR();
 
-  if (IsGPR(Op->Header.Args[0].ID())) {
-    mov(x0, GetReg<RA_64>(Op->Header.Args[0].ID()));
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.PrintValue)));
+  if (IsGPR(Op->Value.ID())) {
+    mov(x0, GetReg<RA_64>(Op->Value.ID()));
+    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintValue)));
   }
   else {
-    fmov(x0, GetSrc(Op->Header.Args[0].ID()).V1D());
+    fmov(x0, GetSrc(Op->Value.ID()).V1D());
     // Bug in vixl that source vector needs to b V1D rather than V2D?
-    fmov(x1, GetSrc(Op->Header.Args[0].ID()).V1D(), 1);
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.PrintVectorValue)));
+    fmov(x1, GetSrc(Op->Value.ID()).V1D(), 1);
+    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintVectorValue)));
   }
   SpillStaticRegs();
   blr(x3);
@@ -219,6 +227,10 @@ DEF_OP(RDRAND) {
   cset(Dst.second, Condition::ne);
 }
 
+DEF_OP(Yield) {
+  hint(SystemHint::YIELD);
+}
+
 #undef DEF_OP
 void Arm64JITCore::RegisterMiscHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
@@ -227,6 +239,7 @@ void Arm64JITCore::RegisterMiscHandlers() {
   REGISTER_OP(CODEBLOCK,  NoOp);
   REGISTER_OP(BEGINBLOCK, NoOp);
   REGISTER_OP(ENDBLOCK,   NoOp);
+  REGISTER_OP(GUESTOPCODE, GuestOpcode);
   REGISTER_OP(FENCE,      Fence);
   REGISTER_OP(BREAK,      Break);
   REGISTER_OP(PHI,        NoOp);
@@ -237,6 +250,7 @@ void Arm64JITCore::RegisterMiscHandlers() {
   REGISTER_OP(INVALIDATEFLAGS,   NoOp);
   REGISTER_OP(PROCESSORID,   ProcessorID);
   REGISTER_OP(RDRAND, RDRAND);
+  REGISTER_OP(YIELD, Yield);
 
 #undef REGISTER_OP
 }

External/FEXCore/Source/Interface/Core/JIT/Arm64/MoveOps.cpp

@@ -15,13 +15,13 @@ DEF_OP(ExtractElementPair) {
   auto Op = IROp->C<IR::IROp_ExtractElementPair>();
   switch (Op->Header.Size) {
     case 4: {
-      auto Src = GetSrcPair<RA_32>(Op->Header.Args[0].ID());
+      auto Src = GetSrcPair<RA_32>(Op->Pair.ID());
       std::array<aarch64::Register, 2> Regs = {Src.first, Src.second};
       mov (GetReg<RA_32>(Node), Regs[Op->Element]);
       break;
     }
     case 8: {
-      auto Src = GetSrcPair<RA_64>(Op->Header.Args[0].ID());
+      auto Src = GetSrcPair<RA_64>(Op->Pair.ID());
       std::array<aarch64::Register, 2> Regs = {Src.first, Src.second};
       mov (GetReg<RA_64>(Node), Regs[Op->Element]);
       break;
@@ -40,15 +40,15 @@ DEF_OP(CreateElementPair) {
   switch (IROp->ElementSize) {
     case 4: {
       Dst = GetSrcPair<RA_32>(Node);
-      RegFirst = GetReg<RA_32>(Op->Header.Args[0].ID());
-      RegSecond = GetReg<RA_32>(Op->Header.Args[1].ID());
+      RegFirst = GetReg<RA_32>(Op->Lower.ID());
+      RegSecond = GetReg<RA_32>(Op->Upper.ID());
       RegTmp = w0;
       break;
     }
     case 8: {
       Dst = GetSrcPair<RA_64>(Node);
-      RegFirst = GetReg<RA_64>(Op->Header.Args[0].ID());
-      RegSecond = GetReg<RA_64>(Op->Header.Args[1].ID());
+      RegFirst = GetReg<RA_64>(Op->Lower.ID());
+      RegSecond = GetReg<RA_64>(Op->Upper.ID());
       RegTmp = x0;
       break;
     }
@@ -70,7 +70,7 @@ DEF_OP(CreateElementPair) {
 
 DEF_OP(Mov) {
   auto Op = IROp->C<IR::IROp_Mov>();
-  mov(GetReg<RA_64>(Node), GetReg<RA_64>(Op->Header.Args[0].ID()));
+  mov(GetReg<RA_64>(Node), GetReg<RA_64>(Op->Value.ID()));
 }
 
 #undef DEF_OP

External/FEXCore/Source/Interface/Core/JIT/JITCore.h

@@ -14,10 +14,13 @@ namespace FEXCore::CPU {
 class CPUBackend;
 
 [[nodiscard]] std::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::Context *ctx,
-                                                           FEXCore::Core::InternalThreadState *Thread,
-                                                           bool CompileThread);
+                                                           FEXCore::Core::InternalThreadState *Thread);
+void InitializeX86JITSignalHandlers(FEXCore::Context::Context *CTX);
+CPUBackendFeatures GetX86JITBackendFeatures();
 
 [[nodiscard]] std::unique_ptr<CPUBackend> CreateArm64JITCore(FEXCore::Context::Context *ctx,
-                                                             FEXCore::Core::InternalThreadState *Thread,
-                                                             bool CompileThread);
+                                                             FEXCore::Core::InternalThreadState *Thread);
+void InitializeArm64JITSignalHandlers(FEXCore::Context::Context *CTX);
+CPUBackendFeatures GetArm64JITBackendFeatures();
+
 } // namespace FEXCore::CPU

External/FEXCore/Source/Interface/Core/JIT/x86_64/BranchOps.cpp

@@ -29,13 +29,6 @@ $end_info$
 
 namespace FEXCore::CPU {
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
 
 DEF_OP(SignalReturn) {
   // Adjust the stack first for a regular return
@@ -43,7 +36,7 @@ DEF_OP(SignalReturn) {
     add(rsp, SpillSlots * 16); // + 8 to consume return address
   }
 
-  jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalReturnHandler)]);
+  jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SignalReturnHandler)]);
 }
 
 DEF_OP(CallbackReturn) {
@@ -53,7 +46,7 @@ DEF_OP(CallbackReturn) {
   }
 
   // Make sure to adjust the refcounter so we don't clear the cache now
-  sub(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalHandlerRefCountPointer)], 1);
+  sub(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter)], 1);
 
   // We need to adjust an additional 8 bytes to get back to the original "misaligned" RSP state
   add(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])], 8);
@@ -91,14 +84,15 @@ DEF_OP(ExitFunction) {
     jmp(qword[rax]);
 
     L(l_BranchHost);
-    dq(ThreadSharedData.Dispatcher->ExitFunctionLinkerAddress);
+    //FEX_TODO(this is not per thread)
+    dq(ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker);
     L(l_BranchGuest);
     dq(NewRIP);
   } else {
     Xbyak::Reg RipReg = GetSrc<RA_64>(Op->NewRIP.ID());
 
     // L1 Cache
-    mov(rcx, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
+    mov(rcx, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer)]);
 
     mov(rax, RipReg);
 
@@ -113,7 +107,7 @@ DEF_OP(ExitFunction) {
 
     L(FullLookup);
     mov(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.rip)], RipReg);
-    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.DispatcherLoopTop)]);
+    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop)]);
   }
 
 #ifdef BLOCKSTATS
@@ -181,13 +175,13 @@ DEF_OP(Syscall) {
   }
 
   mov(rsi, STATE); // Move thread in to rsi
-  mov(rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SyscallHandlerObj)]);
+  mov(rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerObj)]);
   mov(rdx, rsp);
 
   if (NumPush & 1)
     sub(rsp, 8); // Align
   // {rdi, rsi, rdx}
-  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SyscallHandlerFunc)]);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerFunc)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -259,7 +253,7 @@ DEF_OP(ValidateCode) {
   }
 }
 
-DEF_OP(RemoveCodeEntry) {
+DEF_OP(RemoveThreadCodeEntry) {
   auto NumPush = RA64.size();
 
   for (auto &Reg : RA64)
@@ -272,7 +266,7 @@ DEF_OP(RemoveCodeEntry) {
   mov(rax, Entry); // imm64 move
   mov(rsi, rax);
 
-  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.RemoveCodeEntryFromJIT)]);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.RemoveThreadCodeEntryFromJIT)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -296,7 +290,7 @@ DEF_OP(CPUID) {
   // rsi can be in the source registers, so copy argument to edx first
   mov (edx, GetSrc<RA_32>(Op->Header.Args[1].ID()));
   mov (esi, GetSrc<RA_32>(Op->Header.Args[0].ID()));
-  mov (rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.CPUIDObj)]);
+  mov (rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj)]);
 
   auto NumPush = RA64.size();
 
@@ -304,7 +298,7 @@ DEF_OP(CPUID) {
     sub(rsp, 8); // Align
 
   // {rdi, rsi, rdx}
-  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.CPUIDFunction)]);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDFunction)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -320,8 +314,6 @@ DEF_OP(CPUID) {
 #undef DEF_OP
 void X86JITCore::RegisterBranchHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
-  REGISTER_OP(GUESTCALLDIRECT,   GuestCallDirect);
-  REGISTER_OP(GUESTCALLINDIRECT, GuestCallIndirect);
   REGISTER_OP(SIGNALRETURN,      SignalReturn);
   REGISTER_OP(CALLBACKRETURN,    CallbackReturn);
   REGISTER_OP(EXITFUNCTION,      ExitFunction);
@@ -330,7 +322,7 @@ void X86JITCore::RegisterBranchHandlers() {
   REGISTER_OP(SYSCALL,           Syscall);
   REGISTER_OP(THUNK,             Thunk);
   REGISTER_OP(VALIDATECODE,      ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,   RemoveCodeEntry);
+  REGISTER_OP(REMOVETHREADCODEENTRY,   RemoveThreadCodeEntry);
   REGISTER_OP(CPUID,             CPUID);
 #undef REGISTER_OP
 }

External/FEXCore/Source/Interface/Core/JIT/x86_64/EncryptionOps.cpp

@@ -75,16 +75,37 @@ DEF_OP(CRC32) {
   }
 }
 
+DEF_OP(PCLMUL) {
+  auto Op = IROp->C<IR::IROp_PCLMUL>();
+
+  auto Dst = GetDst(Node);
+  auto Src1 = GetSrc(Op->Src1.ID());
+  auto Src2 = GetSrc(Op->Src2.ID());
+
+  switch (Op->Selector) {
+  case 0b00000000:
+  case 0b00000001:
+  case 0b00010000:
+  case 0b00010001:
+    vpclmulqdq(Dst, Src1, Src2, Op->Selector);
+    break;
+  default:
+    LOGMAN_MSG_A_FMT("Unknown PCLMUL selector: {}", Op->Selector);
+    break;
+  }
+}
+
 #undef DEF_OP
 void X86JITCore::RegisterEncryptionHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
-  REGISTER_OP(VAESIMC,     AESImc);
-  REGISTER_OP(VAESENC,     AESEnc);
-  REGISTER_OP(VAESENCLAST, AESEncLast);
-  REGISTER_OP(VAESDEC,     AESDec);
-  REGISTER_OP(VAESDECLAST, AESDecLast);
-  REGISTER_OP(VAESKEYGENASSIST, AESKeyGenAssist);
+  REGISTER_OP(VAESIMC,           AESImc);
+  REGISTER_OP(VAESENC,           AESEnc);
+  REGISTER_OP(VAESENCLAST,       AESEncLast);
+  REGISTER_OP(VAESDEC,           AESDec);
+  REGISTER_OP(VAESDECLAST,       AESDecLast);
+  REGISTER_OP(VAESKEYGENASSIST,  AESKeyGenAssist);
   REGISTER_OP(CRC32,             CRC32);
+  REGISTER_OP(PCLMUL,            PCLMUL);
 #undef REGISTER_OP
 }
 }

External/FEXCore/Source/Interface/Core/JIT/x86_64/JIT.cpp

@@ -43,6 +43,9 @@ $end_info$
 // #define DEBUG_RA 1
 // #define DEBUG_CYCLES
 
+static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
+static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 256;
+
 namespace {
 static void PrintValue(uint64_t Value) {
   LogMan::Msg::DFmt("Value: 0x{:x}", Value);
@@ -55,31 +58,6 @@ static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
 
 namespace FEXCore::CPU {
 
-CodeBuffer AllocateNewCodeBuffer(FEXCore::Context::Context *CTX, size_t Size) {
-  CodeBuffer Buffer;
-  Buffer.Size = Size;
-  Buffer.Ptr = static_cast<uint8_t*>(
-               FEXCore::Allocator::mmap(nullptr,
-                    Buffer.Size,
-                    PROT_READ | PROT_WRITE | PROT_EXEC,
-                    MAP_PRIVATE | MAP_ANONYMOUS,
-                    -1, 0));
-  LOGMAN_THROW_A_FMT(Buffer.Ptr != reinterpret_cast<uint8_t*>(~0ULL), "Couldn't allocate code buffer");
-  if (CTX->Config.GlobalJITNaming()) {
-    CTX->Symbols.RegisterJITSpace(Buffer.Ptr, Buffer.Size);
-  }
-  return Buffer;
-}
-
-void FreeCodeBuffer(CodeBuffer Buffer) {
-  FEXCore::Allocator::munmap(Buffer.Ptr, Buffer.Size);
-}
-
-void X86JITCore::CopyNecessaryDataForCompileThread(CPUBackend *Original) {
-  X86JITCore *Core = reinterpret_cast<X86JITCore*>(Original);
-  ThreadSharedData = Core->ThreadSharedData;
-}
-
 void X86JITCore::PushRegs() {
   sub(rsp, 16 * RAXMM_x.size());
   for (size_t i = 0; i < RAXMM_x.size(); ++i) {
@@ -120,7 +98,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
       case FABI_VOID_U16: {
         PushRegs();
         mov(edi, GetSrc<RA_32>(IROp->Args[0].ID()));
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
         break;
@@ -129,7 +107,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         movss(xmm0, GetSrc(IROp->Args[0].ID()));
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -143,7 +121,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         movsd(xmm0, GetSrc(IROp->Args[0].ID()));
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -158,7 +136,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         mov(edi, GetSrc<RA_32>(IROp->Args[0].ID()));
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -174,7 +152,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -188,7 +166,34 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
+
+        PopRegs();
+
+        movsd(GetDst(Node), xmm0);
+      }
+      break;
+
+      case FABI_F64_F64: {
+        PushRegs();
+
+        movsd(xmm0, GetSrc(IROp->Args[0].ID()));
+
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
+
+        PopRegs();
+
+        movsd(GetDst(Node), xmm0);
+      }
+      break;
+
+      case FABI_F64_F64_F64: {
+        PushRegs();
+
+        movsd(xmm0, GetSrc(IROp->Args[0].ID()));
+        movsd(xmm1, GetSrc(IROp->Args[1].ID()));
+
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -202,7 +207,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -215,7 +220,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -228,7 +233,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -244,7 +249,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdx, GetSrc(IROp->Args[1].ID()));
         pextrq(rcx, GetSrc(IROp->Args[1].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -257,7 +262,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -275,7 +280,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdx, GetSrc(IROp->Args[1].ID()));
         pextrq(rcx, GetSrc(IROp->Args[1].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -295,41 +300,34 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
   }
 }
 
+static uint64_t X86JITCore_ExitFunctionLink(FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
+  auto Thread = Frame->Thread;
+  auto GuestRip = record[1];
+
+  auto HostCode = Thread->LookupCache->FindBlock(GuestRip);
+
+  if (!HostCode) {
+    Thread->CurrentFrame->State.rip = GuestRip;
+    return Frame->Pointers.Common.DispatcherLoopTop;
+  }
+
+  auto LinkerAddress = Frame->Pointers.Common.ExitFunctionLinker;
+  Thread->LookupCache->AddBlockLink(GuestRip, (uintptr_t)record, [record, LinkerAddress]{
+    // undo the link
+    record[0] = LinkerAddress;
+  });
+
+  record[0] = HostCode;
+  return HostCode;
+}
+
 void X86JITCore::Op_NoOp(IR::IROp_Header *IROp, IR::NodeID Node) {
 }
 
-X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, CodeBuffer Buffer, bool CompileThread)
-  : CodeGenerator(Buffer.Size, Buffer.Ptr, nullptr)
-  , CTX {ctx}
-  , ThreadState {Thread}
-  , InitialCodeBuffer {Buffer}
-{
-
-  {
-    // Set up pointers that the JIT needs to load
-    auto &Pointers = ThreadState->CurrentFrame->Pointers.X86;
-    // Process specific
-    Pointers.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
-    Pointers.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
-    Pointers.RemoveCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveCodeEntryFromJit);
-    Pointers.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
-
-    {
-      FEXCore::Utils::MemberFunctionToPointerCast PMF(&FEXCore::CPUIDEmu::RunFunction);
-      Pointers.CPUIDFunction = PMF.GetConvertedPointer();
-    }
-
-    Pointers.SyscallHandlerObj = reinterpret_cast<uint64_t>(CTX->SyscallHandler);
-    Pointers.SyscallHandlerFunc = reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall);
-
-    // Fill in the fallback handlers
-    InterpreterOps::FillFallbackIndexPointers(Pointers.FallbackHandlerPointers);
-
-    // Thread Specific
-    Pointers.SignalHandlerRefCountPointer = reinterpret_cast<uint64_t>(&Dispatcher->SignalHandlerRefCounter);
-  }
-
-  CurrentCodeBuffer = &InitialCodeBuffer;
+X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
+  : CPUBackend(Thread, INITIAL_CODE_SIZE, MAX_CODE_SIZE)
+  , CodeGenerator(0, this, nullptr) // this is not used here
+  , CTX {ctx} {
 
   RAPass = Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA");
 
@@ -358,92 +356,52 @@ X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalTh
   RegisterVectorHandlers();
   RegisterEncryptionHandlers();
 
-  if (!CompileThread) {
-    DispatcherConfig config;
-    config.ExitFunctionLink = reinterpret_cast<uintptr_t>(&ExitFunctionLink);
-    config.ExitFunctionLinkThis = reinterpret_cast<uintptr_t>(this);
+  {
+    auto &Common = ThreadState->CurrentFrame->Pointers.Common;
+    
+    Common.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
+    Common.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
+    Common.RemoveThreadCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveThreadCodeEntryFromJit);
+    Common.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
 
-    Dispatcher = std::make_unique<X86Dispatcher>(CTX, ThreadState, config);
-    DispatchPtr = Dispatcher->DispatchPtr;
-    CallbackPtr = Dispatcher->CallbackPtr;
-
-    ThreadSharedData.SignalHandlerRefCounterPtr = &Dispatcher->SignalHandlerRefCounter;
-    ThreadSharedData.SignalHandlerReturnAddress = Dispatcher->SignalHandlerReturnAddress;
-    ThreadSharedData.UnimplementedInstructionAddress = Dispatcher->UnimplementedInstructionAddress;
-    ThreadSharedData.OverflowExceptionInstructionAddress = Dispatcher->OverflowExceptionInstructionAddress;
-
-    ThreadSharedData.Dispatcher = Dispatcher.get();
-
-    // This will register the host signal handler per thread, which is fine
-    CTX->SignalDelegation->RegisterHostSignalHandler(SIGILL, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      X86JITCore *Core = reinterpret_cast<X86JITCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleSIGILL(Signal, info, ucontext);
-    }, true);
-
-    CTX->SignalDelegation->RegisterHostSignalHandler(SignalDelegator::SIGNAL_FOR_PAUSE, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      X86JITCore *Core = reinterpret_cast<X86JITCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleSignalPause(Signal, info, ucontext);
-    }, true);
-
-    auto GuestSignalHandler = [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) -> bool {
-      X86JITCore *Core = reinterpret_cast<X86JITCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleGuestSignal(Signal, info, ucontext, GuestAction, GuestStack);
-    };
-
-    for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
-      CTX->SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
+    {
+      FEXCore::Utils::MemberFunctionToPointerCast PMF(&FEXCore::CPUIDEmu::RunFunction);
+      Common.CPUIDFunction = PMF.GetConvertedPointer();
     }
+
+    Common.SyscallHandlerObj = reinterpret_cast<uint64_t>(CTX->SyscallHandler);
+    Common.SyscallHandlerFunc = reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall);
+    Common.ExitFunctionLink = reinterpret_cast<uintptr_t>(&X86JITCore_ExitFunctionLink);
+
+    // Fill in the fallback handlers
+    InterpreterOps::FillFallbackIndexPointers(Common.FallbackHandlerPointers);
   }
+
+  // Must be done after Dispatcher init
+  ClearCache();
+}
+
+void X86JITCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
+  CTX->SignalDelegation->RegisterHostSignalHandler(SIGILL, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
+    return Thread->CTX->Dispatcher->HandleSIGILL(Thread, Signal, info, ucontext);
+  }, true);
 }
 
 X86JITCore::~X86JITCore() {
-  for (auto CodeBuffer : CodeBuffers) {
-    FreeCodeBuffer(CodeBuffer);
+
+}
+
+void X86JITCore::EmitDetectionString() {
+  const char JITString[] = "FEXJIT::X86JITCore::";
+  for (char c : JITString) {
+    db(c);
   }
-  CodeBuffers.clear();
-
-
-  FreeCodeBuffer(InitialCodeBuffer);
 }
 
 void X86JITCore::ClearCache() {
-  if (*ThreadSharedData.SignalHandlerRefCounterPtr == 0) {
-    if (!CodeBuffers.empty()) {
-      // If we have more than one code buffer we are tracking then walk them and delete
-      // This is a cleanup step
-      for (auto CodeBuffer : CodeBuffers) {
-        FreeCodeBuffer(CodeBuffer);
-      }
-      CodeBuffers.clear();
-
-      // Set the current code buffer to the initial
-      setNewBuffer(InitialCodeBuffer.Ptr, InitialCodeBuffer.Size);
-      CurrentCodeBuffer = &InitialCodeBuffer;
-    }
-
-    if (CurrentCodeBuffer->Size == MAX_CODE_SIZE) {
-      // Rewind to the start of the code cache start
-      reset();
-    }
-    else {
-      FreeCodeBuffer(InitialCodeBuffer);
-
-      // Resize the code buffer and reallocate our code size
-      CurrentCodeBuffer->Size *= 1.5;
-      CurrentCodeBuffer->Size = std::min(CurrentCodeBuffer->Size, MAX_CODE_SIZE);
-
-      InitialCodeBuffer = AllocateNewCodeBuffer(CTX, CurrentCodeBuffer->Size);
-      setNewBuffer(InitialCodeBuffer.Ptr, InitialCodeBuffer.Size);
-    }
-  }
-  else {
-    // We have signal handlers that have generated code
-    // This means that we can not safely clear the code at this point in time
-    // Allocate some new code buffers that we can switch over to instead
-    auto NewCodeBuffer = AllocateNewCodeBuffer(CTX, X86JITCore::INITIAL_CODE_SIZE);
-    EmplaceNewCodeBuffer(NewCodeBuffer);
-    setNewBuffer(NewCodeBuffer.Ptr, NewCodeBuffer.Size);
-  }
+  auto CodeBuffer = GetEmptyCodeBuffer();
+  setNewBuffer(CodeBuffer->Ptr, CodeBuffer->Size);
+  EmitDetectionString();
 }
 
 IR::PhysicalRegister X86JITCore::GetPhys(IR::NodeID Node) const {
@@ -611,39 +569,27 @@ std::tuple<X86JITCore::SetCC, X86JITCore::CMovCC, X86JITCore::JCC> X86JITCore::G
   return { &CodeGenerator::sete , &CodeGenerator::cmove , &CodeGenerator::je  };
 }
 
-void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData) {
+void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) {
   JumpTargets.clear();
   uint32_t SSACount = IR->GetSSACount();
 
   this->Entry = Entry;
   this->RAData = RAData;
+  this->DebugData = DebugData;
 
   // Fairly excessive buffer range to make sure we don't overflow
-  uint32_t BufferRange = SSACount * 16;
+  uint32_t BufferRange = SSACount * 16 + GDBEnabled * Dispatcher::MaxGDBPauseCheckSize;
   if ((getSize() + BufferRange) > CurrentCodeBuffer->Size) {
-    ThreadState->CTX->ClearCodeCache(ThreadState, false);
+    CTX->ClearCodeCache(ThreadState);
   }
 
-	void *GuestEntry = getCurr<void*>();
+	GuestEntry = getCurr<uint8_t*>();
+  CursorEntry = getSize();
   this->IR = IR;
 
-  if (CTX->GetGdbServerStatus()) {
-    Label RunBlock;
-
-    // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
-    // This happens when single stepping
-    static_assert(sizeof(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-    mov(rax, reinterpret_cast<uint64_t>(CTX));
-
-    // If the value == 0 then branch to the top
-    cmp(dword [rax + (offsetof(FEXCore::Context::Context, Config.RunningMode))], 0);
-    je(RunBlock);
-    // Else we need to pause now
-    mov(rax, ThreadSharedData.Dispatcher->ThreadPauseHandlerAddress);
-    jmp(rax);
-    ud2();
-
-    L(RunBlock);
+  if (GDBEnabled) {
+    auto GDBSize = CTX->Dispatcher->GenerateGDBPauseCheck(GuestEntry, Entry);
+    setSize(getSize() + GDBSize);
   }
 
   LOGMAN_THROW_A_FMT(RAData != nullptr, "Needs RA");
@@ -701,12 +647,13 @@ void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRLi
 
   for (auto [BlockNode, BlockHeader] : IR->GetBlocks()) {
     using namespace FEXCore::IR;
-    {
 #if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
-      auto BlockIROp = BlockHeader->CW<IROp_CodeBlock>();
-      LOGMAN_THROW_A_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");
+    auto BlockIROp = BlockHeader->CW<IROp_CodeBlock>();
+    LOGMAN_THROW_A_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");
 #endif
 
+    auto BlockStartHostCode = getCurr<uint8_t *>();
+    {
       const auto Node = IR->GetID(BlockNode);
       const auto IsTarget = JumpTargets.try_emplace(Node).first;
 
@@ -761,6 +708,13 @@ void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRLi
       OpHandler Handler = OpHandlers[IROp->Op];
       (this->*Handler)(IROp, ID);
     }
+
+    if (DebugData) {
+      DebugData->Subblocks.push_back({
+        static_cast<uint32_t>(BlockStartHostCode - GuestEntry),
+        static_cast<uint32_t>(getCurr<uint8_t *>() - BlockStartHostCode)
+      });
+    }
   }
 
   // Make sure last branch is generated. It certainly can't be eliminated here.
@@ -777,32 +731,22 @@ void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRLi
 
   if (DebugData) {
     DebugData->HostCodeSize = reinterpret_cast<uintptr_t>(GuestExit) - reinterpret_cast<uintptr_t>(GuestEntry);
+    DebugData->Relocations = &Relocations;
   }
+
   return GuestEntry;
 }
 
-uint64_t X86JITCore::ExitFunctionLink(X86JITCore *core, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
-  auto Thread = Frame->Thread;
-  auto GuestRip = record[1];
-
-  auto HostCode = Thread->LookupCache->FindBlock(GuestRip);
-
-  if (!HostCode) {
-    Thread->CurrentFrame->State.rip = GuestRip;
-    return core->ThreadSharedData.Dispatcher->AbsoluteLoopTopAddress;
-  }
-
-  auto LinkerAddress = core->ThreadSharedData.Dispatcher->ExitFunctionLinkerAddress;
-  Thread->LookupCache->AddBlockLink(GuestRip, (uintptr_t)record, [record, LinkerAddress]{
-    // undo the link
-    record[0] = LinkerAddress;
-  });
-
-  record[0] = HostCode;
-  return HostCode;
+std::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
+  return std::make_unique<X86JITCore>(ctx, Thread);
 }
 
-std::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread) {
-  return std::make_unique<X86JITCore>(ctx, Thread, AllocateNewCodeBuffer(ctx, CompileThread ? X86JITCore::MAX_CODE_SIZE : X86JITCore::INITIAL_CODE_SIZE), CompileThread);
+CPUBackendFeatures GetX86JITBackendFeatures() {
+  return CPUBackendFeatures { };
 }
+
+void InitializeX86JITSignalHandlers(FEXCore::Context::Context *CTX) {
+  X86JITCore::InitializeSignalHandlers(CTX);
+}
+
 }

External/FEXCore/Source/Interface/Core/JIT/x86_64/JITClass.h

@@ -6,8 +6,10 @@ $end_info$
 
 #pragma once
 
+#include <FEXCore/IR/RegisterAllocationData.h>
 #include "Interface/Core/BlockSamplingData.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
+#include "Interface/Core/ObjectCache/Relocations.h"
 
 #define XBYAK64
 #include <xbyak/xbyak.h>
@@ -24,13 +26,6 @@ using namespace Xbyak;
 #include <tuple>
 
 namespace FEXCore::CPU {
-struct CodeBuffer {
-  uint8_t *Ptr;
-  size_t Size;
-};
-
-[[nodiscard]] CodeBuffer AllocateNewCodeBuffer(size_t Size);
-void FreeCodeBuffer(CodeBuffer Buffer);
 
 // Temp registers
 // rax, rcx, rdx, rsi, r8, r9,
@@ -57,9 +52,7 @@ const std::array<Xbyak::Xmm, 11> RAXMM_x = {  xmm1, xmm2, xmm3, xmm4, xmm5, xmm6
 class X86JITCore final : public CPUBackend, public Xbyak::CodeGenerator {
 public:
   explicit X86JITCore(FEXCore::Context::Context *ctx,
-                      FEXCore::Core::InternalThreadState *Thread,
-                      CodeBuffer Buffer,
-                      bool CompileThread);
+                      FEXCore::Core::InternalThreadState *Thread);
   ~X86JITCore() override;
 
   [[nodiscard]] std::string GetName() override { return "JIT"; }
@@ -67,7 +60,7 @@ public:
   [[nodiscard]] void *CompileCode(uint64_t Entry,
                                   FEXCore::IR::IRListView const *IR,
                                   FEXCore::Core::DebugData *DebugData,
-                                  FEXCore::IR::RegisterAllocationData *RAData) override;
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;
 
   [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }
 
@@ -75,20 +68,75 @@ public:
 
   void ClearCache() override;
 
-  static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
-  static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 256;
-  void CopyNecessaryDataForCompileThread(CPUBackend *Original) override;
+  static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
 
-  bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true, bool IncludeCompileService = true) const override {
-    return Dispatcher->IsAddressInJITCode(Address, IncludeDispatcher, IncludeCompileService);
-  }
+  void ClearRelocations() override { Relocations.clear(); }
 
 private:
+
+  /**
+   * @name Relocations
+   * @{ */
+    uint64_t GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+    void LoadConstantWithPadding(Xbyak::Reg Reg, uint64_t Constant);
+
+    /**
+     * @brief A literal pair relocation object for named symbol literals
+     */
+    struct NamedSymbolLiteralPair {
+      Label Offset;
+      Relocation MoveABI{};
+    };
+
+    /**
+     * @brief Inserts a thunk relocation
+     *
+     * @param Reg - The GPR to move the thunk handler in to
+     * @param Sum - The hash of the thunk
+     */
+    void InsertNamedThunkRelocation(Xbyak::Reg Reg, const IR::SHA256Sum &Sum);
+
+    /**
+     * @brief Inserts a guest GPR move relocation
+     *
+     * @param Reg - The GPR to move the guest RIP in to
+     * @param Constant - The guest RIP that will be relocated
+     */
+    void InsertGuestRIPMove(Xbyak::Reg Reg, uint64_t Constant);
+
+    /**
+     * @brief Inserts a named symbol as a literal in memory
+     *
+     * Need to use `PlaceNamedSymbolLiteral` with the return value to place the literal in the desired location
+     *
+     * @param Op The named symbol to place
+     *
+     * @return A temporary `NamedSymbolLiteralPair`
+     */
+    NamedSymbolLiteralPair InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+
+    /**
+     * @brief Place the named symbol literal relocation in memory
+     *
+     * @param Lit - Which literal to place
+     */
+
+    void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit);
+
+    std::vector<FEXCore::CPU::Relocation> Relocations;
+
+    ///< Relocation code loading
+    bool ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations);
+
+    /**
+    * @brief Current guest RIP entrypoint
+    */
+    uint64_t CursorEntry{};
+  /**  @} */
+
   Label* PendingTargetLabel{};
   FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *ThreadState;
   FEXCore::IR::IRListView const *IR;
-  std::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
   uint64_t Entry;
 
   std::unordered_map<IR::NodeID, Label> JumpTargets;
@@ -141,41 +189,23 @@ private:
 
   IR::RegisterAllocationPass *RAPass;
   FEXCore::IR::RegisterAllocationData *RAData;
+  FEXCore::Core::DebugData *DebugData;
 
 #ifdef BLOCKSTATS
   bool GetSamplingData {true};
 #endif
 
-  void EmplaceNewCodeBuffer(CodeBuffer Buffer) {
-    CurrentCodeBuffer = &CodeBuffers.emplace_back(Buffer);
-  }
+  static uint64_t ExitFunctionLink(FEXCore::Core::CpuStateFrame *Frame, uint64_t *record);
 
-  static uint64_t ExitFunctionLink(X86JITCore* code, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record);
-
-  // This is the initial code buffer that we will fall back to
-  // In a program without signals and code clearing, we will typically
-  // only have this code buffer
-  CodeBuffer InitialCodeBuffer{};
-  // This is the array of /additional/ code buffers that we may need to allocate
-  // Allocation only occurs when we've hit signals and need to clear code cache
-  // For code safety we can't delete code buffers until outside of all signals
-  std::vector<CodeBuffer> CodeBuffers{};
-
-  // This is the current code buffer that we are tracking
-  CodeBuffer *CurrentCodeBuffer{};
-
-  struct CompilerSharedData {
-    uint64_t SignalHandlerReturnAddress{};
-    uint64_t UnimplementedInstructionAddress{};
-    uint64_t OverflowExceptionInstructionAddress{};
-
-    uint32_t *SignalHandlerRefCounterPtr{};
-    FEXCore::CPU::Dispatcher *Dispatcher{};
-  };
-
-  CompilerSharedData ThreadSharedData;
+  // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
+  void EmitDetectionString();
 
   uint32_t SpillSlots{};
+  /**
+  * @brief Current guest RIP entrypoint
+  */
+  uint8_t *GuestEntry{};
+  
   using SetCC = void (X86JITCore::*)(const Operand& op);
   using CMovCC = void (X86JITCore::*)(const Reg& reg, const Operand& op);
   using JCC = void (X86JITCore::*)(const Label& label, LabelType type);
@@ -274,8 +304,6 @@ private:
   DEF_OP(AtomicFetchNeg);
 
   ///< Branch ops
-  DEF_OP(GuestCallDirect);
-  DEF_OP(GuestCallIndirect);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -284,7 +312,7 @@ private:
   DEF_OP(Syscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveCodeEntry);
+  DEF_OP(RemoveThreadCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops
@@ -319,7 +347,7 @@ private:
   DEF_OP(CacheLineZero);
 
   ///< Misc ops
-  DEF_OP(EndBlock);
+  DEF_OP(GuestOpcode);
   DEF_OP(Fence);
   DEF_OP(Break);
   DEF_OP(Phi);
@@ -329,6 +357,7 @@ private:
   DEF_OP(SetRoundingMode);
   DEF_OP(ProcessorID);
   DEF_OP(RDRAND);
+  DEF_OP(Yield);
 
   ///< Move ops
   DEF_OP(ExtractElementPair);
@@ -434,6 +463,7 @@ private:
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
   DEF_OP(CRC32);
+  DEF_OP(PCLMUL);
 #undef DEF_OP
 };
 

External/FEXCore/Source/Interface/Core/JIT/x86_64/MemoryOps.cpp

@@ -4,6 +4,7 @@ tags: backend|x86-64
 $end_info$
 */
 
+#include "Interface/Core/CPUID.h"
 #include "Interface/Core/JIT/x86_64/JITClass.h"
 
 #include <FEXCore/Core/CoreState.h>

External/FEXCore/Source/Interface/Core/JIT/x86_64/MiscOps.cpp

@@ -7,6 +7,7 @@ $end_info$
 #include "Interface/Context/Context.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/JIT/x86_64/JITClass.h"
+#include "FEXCore/Debug/InternalThreadState.h"
 
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Utils/LogManager.h>
@@ -20,6 +21,12 @@ $end_info$
 namespace FEXCore::CPU {
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 
+DEF_OP(GuestOpcode) {
+  auto Op = IROp->C<IR::IROp_GuestOpcode>();
+  // metadata
+  DebugData->GuestOpcodes.push_back({Op->GuestEntryOffset, getCurr<uint8_t*>() - GuestEntry});
+}
+
 DEF_OP(Fence) {
   auto Op = IROp->C<IR::IROp_Fence>();
   switch (Op->Fence) {
@@ -45,7 +52,7 @@ DEF_OP(Break) {
       break;
     case FEXCore::IR::Break_Overflow: // overflow
       // Need to be outside of JIT cache space to ensure cache clearing correctness
-      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.OverflowExceptionHandler)]);
+      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.OverflowExceptionHandler)]);
       break;
     case FEXCore::IR::Break_Halt: { // HLT
       // Time to quit
@@ -53,7 +60,7 @@ DEF_OP(Break) {
       mov(rsp, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)]);
 
       // Now we need to jump to the thread stop handler
-      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadStopHandler)]);
+      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadStopHandlerSpillSRA)]);
       break;
     }
     case FEXCore::IR::Break_Interrupt3: // INT3
@@ -65,7 +72,7 @@ DEF_OP(Break) {
         }
 
         // This jump target needs to be a constant offset here
-        jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadPauseHandler)]);
+        jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA)]);
       }
       else {
         // If we don't have a gdb server attached then....crash?
@@ -73,7 +80,7 @@ DEF_OP(Break) {
         mov(rsp, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)]);
 
         // Now we need to jump to the thread stop handler
-        jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadStopHandler)]);
+        jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadStopHandlerSpillSRA)]);
       }
     break;
     }
@@ -84,7 +91,7 @@ DEF_OP(Break) {
       }
 
       // Need to be outside of JIT cache space to ensure cache clearing correctness
-      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.UnimplementedInstructionHandler)]);
+      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.UnimplementedInstructionHandler)]);
       break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown Break reason: {}", Op->Reason);
@@ -130,13 +137,13 @@ DEF_OP(Print) {
   PushRegs();
   if (IsGPR(Op->Header.Args[0].ID())) {
     mov (rdi, GetSrc<RA_64>(Op->Header.Args[0].ID()));
-    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.PrintValue)]);
+    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintValue)]);
   }
   else {
     pextrq(rdi, GetSrc(Op->Header.Args[0].ID()), 0);
     pextrq(rsi, GetSrc(Op->Header.Args[0].ID()), 1);
 
-    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.PrintVectorValue)]);
+    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintVectorValue)]);
   }
 
   PopRegs();
@@ -166,6 +173,10 @@ DEF_OP(RDRAND) {
   setc(Dst.second.cvt8());
 }
 
+DEF_OP(Yield) {
+  pause();
+}
+
 #undef DEF_OP
 void X86JITCore::RegisterMiscHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
@@ -174,6 +185,7 @@ void X86JITCore::RegisterMiscHandlers() {
   REGISTER_OP(CODEBLOCK,  NoOp);
   REGISTER_OP(BEGINBLOCK, NoOp);
   REGISTER_OP(ENDBLOCK,   NoOp);
+  REGISTER_OP(GUESTOPCODE, GuestOpcode);
   REGISTER_OP(FENCE,      Fence);
   REGISTER_OP(BREAK,      Break);
   REGISTER_OP(PHI,        NoOp);
@@ -184,6 +196,7 @@ void X86JITCore::RegisterMiscHandlers() {
   REGISTER_OP(INVALIDATEFLAGS,   NoOp);
   REGISTER_OP(PROCESSORID,   ProcessorID);
   REGISTER_OP(RDRAND, RDRAND);
+  REGISTER_OP(YIELD, Yield);
 #undef REGISTER_OP
 }
 }

External/FEXCore/Source/Interface/Core/JIT/x86_64/x64Relocations.cpp

@@ -0,0 +1,139 @@
+/*
+$info$
+tags: backend|x86-64
+desc: relocation logic of the x86-64 splatter backend
+$end_info$
+*/
+#include "Interface/Context/Context.h"
+#include "Interface/Core/JIT/x86_64/JITClass.h"
+#include "Interface/HLE/Thunks/Thunks.h"
+
+namespace FEXCore::CPU {
+uint64_t X86JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+  switch (Op) {
+    case FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER:
+      return ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker;
+    break;
+    default:
+      ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
+    break;
+  }
+  return ~0ULL;
+
+}
+
+void X86JITCore::LoadConstantWithPadding(Xbyak::Reg Reg, uint64_t Constant) {
+  // The maximum size a move constant can be in bytes
+  // Need to NOP pad to this size to ensure backpatching is always the same size
+  // Calculated as:
+  // [Rex]
+  // [Mov op]
+  // [8 byte constant]
+  //
+  // All other move types are smaller than this. xbyak will use a NOP slide which is quite quick
+  constexpr static size_t MAX_MOVE_SIZE = 10;
+  auto StartingOffset = getSize();
+  mov(Reg, Constant);
+  auto MoveSize = getSize() - StartingOffset;
+  auto NOPPadSize = MAX_MOVE_SIZE - MoveSize;
+  nop(NOPPadSize);
+}
+
+X86JITCore::NamedSymbolLiteralPair X86JITCore::InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+  NamedSymbolLiteralPair Lit {
+    .MoveABI = {
+      .NamedSymbolLiteral = {
+        .Header = {
+          .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL,
+        },
+        .Symbol = Op,
+        .Offset = 0,
+      },
+    },
+  };
+  return Lit;
+}
+
+void X86JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit) {
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = getSize();
+  Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - CursorEntry;
+
+  uint64_t Pointer = GetNamedSymbolLiteral(Lit.MoveABI.NamedSymbolLiteral.Symbol);
+
+  L(Lit.Offset);
+  dq(Pointer);
+  Relocations.emplace_back(Lit.MoveABI);
+}
+
+
+void X86JITCore::InsertGuestRIPMove(Xbyak::Reg Reg, uint64_t Constant) {
+  Relocation MoveABI{};
+  MoveABI.GuestRIPMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE;
+
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = getSize();
+  MoveABI.GuestRIPMove.Offset = CurrentCursor - CursorEntry;
+  MoveABI.GuestRIPMove.GuestRIP = Constant;
+  MoveABI.GuestRIPMove.RegisterIndex = Reg.getIdx();
+
+  if (CTX->Config.CacheObjectCodeCompilation()) {
+    LoadConstantWithPadding(Reg, Constant);
+  }
+  else {
+    mov(Reg, Constant);
+  }
+
+  Relocations.emplace_back(MoveABI);
+}
+
+bool X86JITCore::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations) {
+  size_t DataIndex{};
+  for (size_t j = 0; j < NumRelocations; ++j) {
+    const FEXCore::CPU::Relocation *Reloc = reinterpret_cast<const FEXCore::CPU::Relocation *>(&EntryRelocations[DataIndex]);
+    LOGMAN_THROW_A_FMT((DataIndex % alignof(Relocation)) == 0, "Alignment of relocation wasn't adhered to");
+
+    switch (Reloc->Header.Type) {
+      case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
+        uint64_t Pointer = GetNamedSymbolLiteral(Reloc->NamedSymbolLiteral.Symbol);
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        setSize(CursorEntry + Reloc->NamedSymbolLiteral.Offset);
+
+        // Place the pointer
+        dq(Pointer);
+
+        DataIndex += sizeof(Reloc->NamedSymbolLiteral);
+        break;
+      }
+      case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
+        uint64_t Pointer = reinterpret_cast<uint64_t>(CTX->ThunkHandler->LookupThunk(Reloc->NamedThunkMove.Symbol));
+        if (Pointer == ~0ULL) {
+          return false;
+        }
+
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        setSize(CursorEntry + Reloc->NamedThunkMove.Offset);
+        LoadConstantWithPadding(Xbyak::Reg64(Reloc->NamedThunkMove.RegisterIndex), Pointer);
+        DataIndex += sizeof(Reloc->NamedThunkMove);
+        break;
+      }
+      case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE:
+        // XXX: Reenable once the JIT Object Cache is upstream
+        // XXX: Should spin the relocation list, create a list of guest RIP moves, and ask for them all once, reduces lock contention.
+        uint64_t Pointer = ~0ULL; // EmitterCTX->JITObjectCache->FindRelocatedRIP(Reloc->GuestRIPMove.GuestRIP);
+        if (Pointer == ~0ULL) {
+          return false;
+        }
+
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        setSize(CursorEntry + Reloc->GuestRIPMove.Offset);
+        LoadConstantWithPadding(Xbyak::Reg64(Reloc->GuestRIPMove.RegisterIndex), Pointer);
+        DataIndex += sizeof(Reloc->GuestRIPMove);
+        break;
+    }
+  }
+
+  return true;
+}
+}
+

External/FEXCore/Source/Interface/Core/LookupCache.cpp

@@ -52,15 +52,8 @@ LookupCache::~LookupCache() {
   FEXCore::Allocator::munmap(reinterpret_cast<void*>(L1Pointer), L1_SIZE);
 }
 
-void LookupCache::HintUsedRange(uint64_t Address, uint64_t Size) {
-  // Tell the kernel we will definitely need [Address, Address+Size) mapped for the page pointer
-  // Page Pointer is allocated per page, so shift by page size
-  Address >>= 12;
-  Size >>= 12;
-  madvise(reinterpret_cast<void*>(PagePointer + Address), Size, MADV_WILLNEED);
-}
-
 void LookupCache::ClearL2Cache() {
+  std::lock_guard<std::recursive_mutex> lk(WriteLock);
   // Clear out the page memory
   madvise(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8, MADV_DONTNEED);
   madvise(reinterpret_cast<void*>(PageMemory), CODE_SIZE, MADV_DONTNEED);
@@ -68,6 +61,8 @@ void LookupCache::ClearL2Cache() {
 }
 
 void LookupCache::ClearCache() {
+  std::lock_guard<std::recursive_mutex> lk(WriteLock);
+
   // Clear L1
   madvise(reinterpret_cast<void*>(L1Pointer), L1_SIZE, MADV_DONTNEED);
   // Clear L2

External/FEXCore/Source/Interface/Core/LookupCache.h

@@ -7,6 +7,7 @@
 #include <stddef.h>
 #include <utility>
 #include <vector>
+#include <mutex>
 
 namespace FEXCore {
 namespace Context {
@@ -24,38 +25,73 @@ public:
   LookupCache(FEXCore::Context::Context *CTX);
   ~LookupCache();
 
-  using LookupCacheIter = uintptr_t;
-  uintptr_t End() { return 0; }
-
   uintptr_t FindBlock(uint64_t Address) {
-    auto HostCode = FindCodePointerForAddress(Address);
-    if (HostCode) {
-      return HostCode;
-    } else {
-      auto HostCode = BlockList.find(Address);
+    // Try L1, no lock needed
+    auto &L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    if (L1Entry.GuestCode == Address) {
+      return L1Entry.HostCode;
+    }
 
-      if (HostCode != BlockList.end()) {
-        CacheBlockMapping(Address, HostCode->second);
-        return HostCode->second;
-      } else {
-        return 0;
+    // L2 and L3 need to be locked
+    std::lock_guard<std::recursive_mutex> lk(WriteLock);
+
+    // Try L2
+    const auto PageIndex = (Address & (VirtualMemSize -1)) >> 12;
+    const auto PageOffset = Address & (0x0FFF);
+
+    const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+    auto LocalPagePointer = Pointers[PageIndex];
+
+    // Do we a page pointer for this address?
+    if (LocalPagePointer) {
+      // Find there pointer for the address in the blocks
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+
+      if (BlockPointers[PageOffset].GuestCode == Address)
+      {
+        L1Entry.GuestCode = Address;
+        L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
+        return L1Entry.HostCode;
       }
     }
+    
+    // Try L3
+    auto HostCode = BlockList.find(Address);
+
+    if (HostCode != BlockList.end()) {
+      CacheBlockMapping(Address, HostCode->second);
+      return HostCode->second;
+    }
+      
+    // Failed to find
+    return 0;
   }
 
   std::map<uint64_t, std::vector<uint64_t>> CodePages;
 
-  void AddBlockMapping(uint64_t Address, void *HostCode, uint64_t Start, uint64_t Length) { 
-#if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
-    auto InsertPoint =
-#endif
-    BlockList.emplace(Address, (uintptr_t)HostCode);
-    LOGMAN_THROW_A_FMT(InsertPoint.second == true, "Dupplicate block mapping added");
+  // Appends Block {Address} to CodePages [Start, Start + Length)
+  // Returns true if new pages are marked as containing code
+  bool AddBlockExecutableRange(uint64_t Address, uint64_t Start, uint64_t Length) {
+    std::lock_guard<std::recursive_mutex> lk(WriteLock);
+    
+    bool rv = false;
 
-    for (auto CurrentPage = Start >> 12, EndPage = (Start + Length) >> 12; CurrentPage <= EndPage; CurrentPage++) {
-      CodePages[CurrentPage].push_back(Address);
+    for (auto CurrentPage = Start >> 12, EndPage = (Start + Length -1) >> 12; CurrentPage <= EndPage; CurrentPage++) {
+      auto &CodePage = CodePages[CurrentPage];
+      rv |= CodePage.size() == 0;
+      CodePage.push_back(Address);
     }
 
+    return rv;
+  }
+
+  // Adds to Guest -> Host code mapping
+  void AddBlockMapping(uint64_t Address, void *HostCode) {
+    std::lock_guard<std::recursive_mutex> lk(WriteLock);
+    
+    [[maybe_unused]] auto Inserted = BlockList.emplace(Address, (uintptr_t)HostCode).second;
+    LOGMAN_THROW_A_FMT(Inserted, "Duplicate block mapping added");
+
     // There is no need to update L1 or L2, they will get updated on first lookup
     // However, adding to L1 here increases performance
     auto &L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
@@ -65,6 +101,8 @@ public:
 
   void Erase(uint64_t Address) {
 
+    std::lock_guard<std::recursive_mutex> lk(WriteLock);
+
     // Sever any links to this block
     auto lower = BlockLinks.lower_bound({Address, 0});
     auto upper = BlockLinks.upper_bound({Address, UINTPTR_MAX});
@@ -78,7 +116,10 @@ public:
     // Do L1
     auto &L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
     if (L1Entry.GuestCode == Address) {
-      L1Entry.GuestCode = L1Entry.HostCode = 0;
+      L1Entry.GuestCode = 0;
+      // Leave L1Entry.HostCode as is, so that concurrent lookups won't read a null pointer
+      // This is a soft guarantee for cross thread invalidation, as atomics are not used
+      // and it hasn't been thoroughly tested
     }
 
     // Do full map
@@ -101,14 +142,14 @@ public:
 
 
   void AddBlockLink(uint64_t GuestDestination, uintptr_t HostLink, const std::function<void()> &delinker) {
+    std::lock_guard<std::recursive_mutex> lk(WriteLock);
+
     BlockLinks.insert({{GuestDestination, HostLink}, delinker});
   }
 
   void ClearCache();
   void ClearL2Cache();
 
-  void HintUsedRange(uint64_t Address, uint64_t Size);
-
   uintptr_t GetL1Pointer() const { return L1Pointer; }
   uintptr_t GetPagePointer() const { return PagePointer; }
   uintptr_t GetVirtualMemorySize() const { return VirtualMemSize; }
@@ -116,8 +157,19 @@ public:
   constexpr static size_t L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
   constexpr static size_t L1_ENTRIES_MASK = L1_ENTRIES - 1;
 
+  // This needs to be taken before reads or writes to L2, L3, CodePages, Thread::DebugStore,
+  // and before writes to L1. Concurrent access from a thread that this LookupCache doesn't belong to 
+  // may only happen during cross thread invalidation (::Erase).
+  // All other operations must be done from the owning thread.
+  // Some care is taken so that L1 lookups can be done without locks, and even tearing is unlikely to lead to a crash.
+  // This approach has not been fully vetted yet.
+  // Also note that L1 lookups might be inlined in the JIT Dispatcher and/or block ends.
+  std::recursive_mutex WriteLock;
+
 private:
   void CacheBlockMapping(uint64_t Address, uintptr_t HostCode) { 
+    std::lock_guard<std::recursive_mutex> lk(WriteLock);
+
     // Do L1
     auto &L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
     L1Entry.GuestCode = Address;
@@ -167,38 +219,6 @@ private:
     return PageMemory + NewBase;
   }
 
-  uintptr_t FindCodePointerForAddress(uint64_t Address) {
-    
-    // Do L1
-    auto &L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
-    if (L1Entry.GuestCode == Address) {
-      return L1Entry.HostCode;
-    }
-
-    auto FullAddress = Address;
-    Address = Address & (VirtualMemSize -1);
-
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
-    uintptr_t *Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // We don't have a page pointer for this address
-      return 0;
-    }
-
-    // Find there pointer for the address in the blocks
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
-
-    if (BlockPointers[PageOffset].GuestCode == FullAddress)
-    {
-      L1Entry.GuestCode = FullAddress;
-      return L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
-    }
-    else
-      return 0;
-  }
-
   uintptr_t PagePointer;
   uintptr_t PageMemory;
   uintptr_t L1Pointer;

External/FEXCore/Source/Interface/Core/ObjectCache/CodeObjectSerializationConfig.h

@@ -0,0 +1,84 @@
+#pragma once
+
+#include <cstdint>
+
+namespace FEXCore::CodeSerialize {
+  // If any of the config options mismatch on load then the cache won't be used
+  // Any of these will result in codegen changes
+  struct CodeObjectSerializationConfig {
+		// Cookie in the header of the file, isn't part of the config hash
+    uint64_t Cookie{};
+
+    // Instructions per block configuration
+    int32_t MaxInstPerBlock{};
+
+    // Follows CPUID 4000_0001_EAX[3:0]
+    unsigned Arch : 4;
+
+    // Multiblock enabled
+    bool MultiBlock : 1;
+
+    // TSO enabled
+    bool TSOEnabled : 1;
+
+    // ABI local flag unsafe optimization
+    bool ABILocalFlags : 1;
+
+    // ABI no PF unsafe optimization
+    bool ABINoPF : 1;
+
+    // Static register allocation enabled
+    bool SRA : 1;
+
+    // Paranoid TSO mode enabled
+    bool ParanoidTSO : 1;
+
+    // Guest code execution mode (We don't support live mode switch)
+    bool Is64BitMode : 1;
+
+    // SMC checks style
+    unsigned SMCChecks : 2;
+
+    // x87 reduced precision
+    bool x87ReducedPrecision : 1;
+
+    // Padding to remove uninitialized data warning from asan
+    // Shows remaining amount of bits available for config
+    unsigned _Pad : 18;
+
+    bool operator==(CodeObjectSerializationConfig const &other) const {
+      return Cookie == other.Cookie &&
+        MaxInstPerBlock == other.MaxInstPerBlock &&
+        Arch == other.Arch &&
+        MultiBlock == other.MultiBlock &&
+        TSOEnabled == other.TSOEnabled &&
+        ABILocalFlags == other.ABILocalFlags &&
+        ABINoPF == other.ABINoPF &&
+        SRA == other.SRA &&
+        ParanoidTSO == other.ParanoidTSO &&
+        Is64BitMode == other.Is64BitMode &&
+        SMCChecks == other.SMCChecks &&
+        x87ReducedPrecision == other.x87ReducedPrecision;
+    }
+    static uint64_t GetHash(CodeObjectSerializationConfig const &other) {
+      // For < 64-bits of data just pack directly
+      // Skip the cookie
+      uint64_t Hash{};
+      Hash <<= 32; Hash |= other.MaxInstPerBlock;
+      Hash <<= 1;  Hash |= other.Arch;
+      Hash <<= 1;  Hash |= other.MultiBlock;
+      Hash <<= 1;  Hash |= other.TSOEnabled;
+      Hash <<= 1;  Hash |= other.ABILocalFlags;
+      Hash <<= 1;  Hash |= other.ABINoPF;
+      Hash <<= 1;  Hash |= other.SRA;
+      Hash <<= 1;  Hash |= other.ParanoidTSO;
+      Hash <<= 1;  Hash |= other.Is64BitMode;
+      Hash <<= 2;  Hash |= other.SMCChecks;
+      Hash <<= 1;  Hash |= other.x87ReducedPrecision;
+      return Hash;
+    }
+  };
+
+	static_assert(sizeof(CodeObjectSerializationConfig) == 16, "Size changed");
+  static_assert((sizeof(CodeObjectSerializationConfig) - sizeof(uint64_t)) == 8, "Config size exceeded 64its. Need to change how the hash is generated!");
+}

External/FEXCore/Source/Interface/Core/ObjectCache/JobHandling.cpp

@@ -0,0 +1,127 @@
+#include "Interface/Context/Context.h"
+#include "Interface/Core/ObjectCache/ObjectCacheService.h"
+
+#include <FEXCore/Config/Config.h>
+
+#include <fcntl.h>
+#include <filesystem>
+#include <memory>
+#include <string>
+#include <sys/uio.h>
+#include <sys/mman.h>
+#include <xxhash.h>
+
+namespace FEXCore::CodeSerialize {
+  void AsyncJobHandler::AsyncAddNamedRegionJob(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename) {
+    // This function adds a named region *JOB* to our named region handler
+    // This needs to be as fast as possible to keep out of the way of the JIT
+
+    auto BaseFilename = std::filesystem::path(filename).filename().string();
+
+    if (!BaseFilename.empty()) {
+      // Create a new entry that once set up will be put in to our section object map
+      auto Entry = std::make_unique<CodeRegionEntry>(
+        Base,
+        Size,
+        Offset,
+        filename,
+        NamedRegionHandler->DefaultCodeHeader(Base, Offset)
+      );
+
+      // Lock the job ref counter so we can block anything attempting to use the entry before it is loaded
+      Entry->NamedJobRefCountMutex.lock();
+
+      CodeRegionMapType::iterator EntryIterator;
+      {
+        std::unique_lock lk {CodeObjectCacheService->GetEntryMapMutex()};
+
+        auto &EntryMap = CodeObjectCacheService->GetEntryMap();
+
+        auto it = EntryMap.emplace(Base, std::move(Entry));
+        if (!it.second) {
+          // This happens when an application overwrites a previous region without unmapping what was there
+
+          // Lock this entry's Named job reference counter.
+          // Once this passes then we know that this section has been loaded.
+          it.first->second->NamedJobRefCountMutex.lock();
+
+          // Finalize anything the region needs to do first.
+          CodeObjectCacheService->DoCodeRegionClosure(it.first->second->Base, it.first->second.get());
+
+          // munmap the file that was mapped
+          FEXCore::Allocator::munmap(it.first->second->CodeData, it.first->second->FileSize);
+
+          // Remove this entry from the unrelocated map as well
+          {
+            std::unique_lock lk2 {CodeObjectCacheService->GetUnrelocatedEntryMapMutex()};
+            CodeObjectCacheService->GetUnrelocatedEntryMap().erase(it.first->second->EntryHeader.OriginalBase);
+          }
+
+          // Now overwrite the entry in the map
+          it = EntryMap.insert_or_assign(Base, std::move(Entry));
+          EntryIterator = it.first;
+        }
+        else {
+          // No overwrite, just insert
+          EntryIterator = it.first;
+        }
+      }
+
+      // Now that this entry has been added to the map, we can insert a load job using the entry iterator.
+      // This allows us to quickly unblock the JIT thread when it is loading multiple regions and have the async thread
+      // do the loading for us.
+      //
+      // Create the async work queue job now so it can load
+      NamedRegionHandler->AsyncAddNamedRegionWorkItem(BaseFilename, filename, true, EntryIterator);
+
+      // Tell the async thread that it has work to do
+      CodeObjectCacheService->NotifyWork();
+    }
+  }
+
+  void AsyncJobHandler::AsyncRemoveNamedRegionJob(uintptr_t Base, uintptr_t Size) {
+    // Removing a named region through the job system
+    // We need to find the entry that we are deleting first
+    std::unique_ptr<CodeRegionEntry> EntryPointer;
+    {
+      std::unique_lock lk {CodeObjectCacheService->GetEntryMapMutex()};
+
+      auto &EntryMap = CodeObjectCacheService->GetEntryMap();
+      auto it = EntryMap.find(Base);
+      if (it != EntryMap.end()) {
+        // Lock the job ref counter since we are erasing it
+        // Once this passes it will have been loaded
+        it->second->NamedJobRefCountMutex.lock();
+
+        // Take the pointer from the map
+        EntryPointer = std::move(it->second);
+
+        // We can now unmap the file data
+        FEXCore::Allocator::munmap(EntryPointer->CodeData, EntryPointer->FileSize);
+
+        // Remove this from the entry map
+        EntryMap.erase(it);
+
+        // Remove this entry from the unrelocated map as well
+        {
+          std::unique_lock lk2 {CodeObjectCacheService->GetUnrelocatedEntryMapMutex()};
+          CodeObjectCacheService->GetUnrelocatedEntryMap().erase(EntryPointer->EntryHeader.OriginalBase);
+        }
+      }
+      else {
+        // Tried to remove something that wasn't in our code object tracking
+        return;
+      }
+
+      // Create the async work queue job now so it can finalize what it needs to do
+      NamedRegionHandler->AsyncRemoveNamedRegionWorkItem(Base, Size, std::move(EntryPointer));
+
+      // Tell the async thread that it has work to do
+      CodeObjectCacheService->NotifyWork();
+    }
+  }
+
+  void AsyncJobHandler::AsyncAddSerializationJob(std::unique_ptr<SerializationJobData> Data) {
+    // XXX: Actually add serialization job
+  }
+}

External/FEXCore/Source/Interface/Core/ObjectCache/NamedRegionObjectHandler.cpp

@@ -0,0 +1,71 @@
+#include "Interface/Context/Context.h"
+#include "Interface/Core/ObjectCache/ObjectCacheService.h"
+
+#include <FEXCore/Config/Config.h>
+
+namespace FEXCore::CodeSerialize {
+  NamedRegionObjectHandler::NamedRegionObjectHandler(FEXCore::Context::Context *ctx) {
+    DefaultSerializationConfig.Cookie = CODE_COOKIE;
+
+    // Initialize the Arch from CPUID
+    uint32_t Arch = ctx->CPUID.RunFunction(0x4000'0001, 0).eax & 0xF;
+    DefaultSerializationConfig.Arch = Arch;
+
+    DefaultSerializationConfig.MaxInstPerBlock = ctx->Config.MaxInstPerBlock;
+    DefaultSerializationConfig.MultiBlock = ctx->Config.Multiblock;
+    DefaultSerializationConfig.TSOEnabled = ctx->Config.TSOEnabled;
+    DefaultSerializationConfig.ABILocalFlags = ctx->Config.ABILocalFlags;
+    DefaultSerializationConfig.ABINoPF = ctx->Config.ABINoPF;
+    DefaultSerializationConfig.SRA = ctx->Config.StaticRegisterAllocation;
+    DefaultSerializationConfig.ParanoidTSO = ctx->Config.ParanoidTSO;
+    DefaultSerializationConfig.Is64BitMode = ctx->Config.Is64BitMode;
+    DefaultSerializationConfig.SMCChecks = ctx->Config.SMCChecks;
+    DefaultSerializationConfig.x87ReducedPrecision = ctx->Config.x87ReducedPrecision;
+  }
+
+  void NamedRegionObjectHandler::AddNamedRegionObject(CodeRegionMapType::iterator Entry, const std::string &base_filename, const std::string &filename, bool Executable) {
+    // XXX: Add named region objects
+
+    // XXX: Until entry loading is complete just claim it is loaded
+    Entry->second->NamedJobRefCountMutex.unlock();
+  }
+
+  void NamedRegionObjectHandler::RemoveNamedRegionObject(uintptr_t Base, uintptr_t Size, std::unique_ptr<CodeRegionEntry> Entry) {
+    // XXX: Remove named region objects
+
+    // XXX: Until entry loading is complete just claim it is loaded
+    Entry->NamedJobRefCountMutex.unlock();
+  }
+
+  void NamedRegionObjectHandler::HandleNamedRegionObjectJobs() {
+    // Walk through all of our jobs sequentially until the work queue is empty
+    while (NamedWorkQueueJobs.load()) {
+      std::unique_ptr<AsyncJobHandler::NamedRegionWorkItem> WorkItem;
+
+      {
+        // Lock the work queue mutex for a short moment and grab an item from the list
+        std::unique_lock lk {NamedWorkQueueMutex};
+        size_t WorkItems = WorkQueue.size();
+        if (WorkItems != 0) {
+          WorkItem = std::move(WorkQueue.front());
+          WorkQueue.pop();
+        }
+
+        // Atomically update the number of jobs
+        --NamedWorkQueueJobs;
+      }
+
+      if (WorkItem) {
+        if (WorkItem->GetType() == AsyncJobHandler::NamedRegionJobType::JOB_ADD_NAMED_REGION) {
+          auto WorkAdd = static_cast<AsyncJobHandler::WorkItemAddNamedRegion *>(WorkItem.get());
+          AddNamedRegionObject(WorkAdd->Entry, WorkAdd->BaseFilename, WorkAdd->Filename, WorkAdd->Executable);
+        }
+
+        if (WorkItem->GetType() == AsyncJobHandler::NamedRegionJobType::JOB_REMOVE_NAMED_REGION) {
+          auto WorkRemove = static_cast<AsyncJobHandler::WorkItemRemoveNamedRegion *>(WorkItem.get());
+          RemoveNamedRegionObject(WorkRemove->Base, WorkRemove->Size, std::move(WorkRemove->Entry));
+        }
+      }
+    }
+  }
+}

External/FEXCore/Source/Interface/Core/ObjectCache/ObjectCacheService.cpp

@@ -0,0 +1,85 @@
+#include "Interface/Core/ObjectCache/ObjectCacheService.h"
+
+#include <FEXCore/Config/Config.h>
+
+#include <memory>
+
+namespace {
+  static void* ThreadHandler(void *Arg) {
+    FEXCore::CodeSerialize::CodeObjectSerializeService *This = reinterpret_cast<FEXCore::CodeSerialize::CodeObjectSerializeService*>(Arg);
+    This->ExecutionThread();
+    return nullptr;
+  }
+}
+
+namespace FEXCore::CodeSerialize {
+  CodeObjectSerializeService::CodeObjectSerializeService(FEXCore::Context::Context *ctx)
+    : CTX {ctx}
+    , AsyncHandler { &NamedRegionHandler , this }
+    , NamedRegionHandler { ctx } {
+    Initialize();
+  }
+
+  void CodeObjectSerializeService::Shutdown() {
+    if (CTX->Config.CacheObjectCodeCompilation() == FEXCore::Config::ConfigObjectCodeHandler::CONFIG_NONE) {
+      return;
+    }
+
+    WorkerThreadShuttingDown = true;
+
+    // Kick the working thread
+    WorkAvailable.NotifyAll();
+
+    if (WorkerThread->joinable()) {
+      // Wait for worker thread to close down
+      WorkerThread->join(nullptr);
+    }
+  }
+
+  void CodeObjectSerializeService::Initialize() {
+    // Add a canary so we don't crash on empty map iterator handling
+    auto it = AddressToEntryMap.insert_or_assign(~0ULL, std::make_unique<CodeRegionEntry>());
+		UnrelocatedAddressToEntryMap.insert_or_assign(~0ULL, it.first->second.get());
+
+		uint64_t OldMask = FEXCore::Threads::SetSignalMask(~0ULL);
+		WorkerThread = FEXCore::Threads::Thread::Create(ThreadHandler, this);
+    FEXCore::Threads::SetSignalMask(OldMask);
+	}
+
+  void CodeObjectSerializeService::DoCodeRegionClosure(uint64_t Base, CodeRegionEntry *it) {
+    if (Base == ~0ULL) {
+      // Don't do closure on canary
+      return;
+    }
+    // XXX: Do code region closure
+  }
+
+  CodeObjectFileSection const *CodeObjectSerializeService::FetchCodeObjectFromCache(uint64_t GuestRIP) {
+    // XXX: Actually fetch code objects from cache
+    return nullptr;
+  }
+
+  void CodeObjectSerializeService::ExecutionThread() {
+    // Set our thread name so we can see its relation
+    char ThreadName[16] = "ObjectCodeSeri\0";
+    pthread_setname_np(pthread_self(), ThreadName);
+    while (WorkerThreadShuttingDown.load() != true) {
+      // Wait for work
+      WorkAvailable.Wait();
+
+      // Handle named region async jobs first. Highest priority
+      NamedRegionHandler.HandleNamedRegionObjectJobs();
+
+      // XXX: Handle code serialization jobs second.
+    }
+
+    // Do final code region closures on thread shutdown
+    for (auto &it : AddressToEntryMap) {
+      DoCodeRegionClosure(it.first, it.second.get());
+    }
+
+    // Safely clear our maps now
+    AddressToEntryMap.clear();
+    UnrelocatedAddressToEntryMap.clear();
+	}
+}

External/FEXCore/Source/Interface/Core/ObjectCache/ObjectCacheService.h

@@ -0,0 +1,459 @@
+#pragma once
+#include "Interface/Context/Context.h"
+#include "Interface/Core/ObjectCache/Relocations.h"
+#include "Interface/Core/ObjectCache/CodeObjectSerializationConfig.h"
+#include "Interface/IR/AOTIR.h"
+
+#include <FEXCore/Utils/Event.h>
+#include <FEXCore/Utils/Threads.h>
+
+#include <map>
+#include <memory>
+#include <shared_mutex>
+#include <string>
+#include <vector>
+#include <tsl/robin_map.h>
+
+namespace FEXCore::CodeSerialize {
+  // XXX: Does this need to be signal safe?
+  using CodeSerializationMutex = std::shared_mutex;
+  struct CodeSerializationData {
+  };
+
+  struct CodeObjectFileSection {
+    bool Serialized;
+    bool Invalid;
+    const CodeSerializationData *Data;
+    const char *HostCode;
+    uint64_t NumRelocations;
+    const char *Relocations;
+  };
+
+  /**
+   * @brief This is the file header that lives at the start of an object cache file
+   *
+   * This header is updated from multiple processes!
+   * Care must be taken to use OS locks when updating the file backing including this header
+   */
+  struct CodeObjectSerializationHeader {
+    // The configuration that this file has
+    CodeObjectSerializationConfig Config;
+    // The original RIP that this object section was mapped at
+    uint64_t OriginalBase{};
+    // The original offset in to the file that this object section was loaded from
+    uint64_t OriginalOffset{};
+    // Total amount of code that should be in this file
+    uint64_t TotalCodeSize{};
+    // Used to reserve the TSL map
+    uint64_t NumCodeEntries{};
+    // The number of relocations that point to this section
+    uint64_t NumRelocationsTo{};
+    // Total relocations in this file
+    uint64_t TotalRelocationsCount{};
+  };
+
+  struct CodeRegionEntry {
+    /**
+     * @name Threaded initialization objects for the initial object creation
+     * @{ */
+      // Base address in memory where the code region is at
+      uint64_t Base{};
+
+      // Size of this code entry
+      uint64_t Size{};
+
+      // The offset inside the file that is mapped to Base
+      uint64_t Offset{};
+
+      // Filename of the object
+      std::string Filename{};
+
+      CodeObjectSerializationHeader EntryHeader{};
+    /**  @} */
+
+    // The filename of the object cache for this entry
+    std::string ObjectEntrySourceFilename{};
+
+    // In the case of file corruption that we can detect, we can disable serialization early for an entry
+    // We should be resiliant to corruption but things happen
+    bool StillSerializing {true};
+
+    // Long lived FD for serialization if we have multiple jobs to serialize
+    // Bursts of code entries are common and this reduces file lock overhead
+    //
+    // Especially useful over network mounts where file locks are very slow
+    int CurrentSerializedFD {-1};
+
+    /**
+     * @name Objects required to sync objects between threads
+     * @{ */
+      // Refcount for the number of outstanding code entries waiting to be written for this object section
+      CodeSerializationMutex ObjectJobRefCountMutex;
+
+      // Refcount for outstanding named object region entry loading itself
+      // Will block JIT code cache look up when this has a unique_lock held
+      CodeSerializationMutex NamedJobRefCountMutex;
+    /**  @} */
+
+    /**
+     * @name Object Entry data management
+     * @{ */
+
+      /**
+       * @name This is the raw file data that we loaded from the code region entry file
+       * @{ */
+        char *CodeData{};
+        size_t FileSize{};
+
+        std::vector<CodeObjectFileSection> FileCodeSections;
+      /**  @} */
+
+      // This per section map takes the most time to load and needs to be quick
+      // This is the map of all code segments for this entry
+      tsl::robin_map<uint64_t, CodeObjectFileSection*> SectionLookupMap{};
+    /**  @} */
+
+    // Default initialization
+    CodeRegionEntry() = default;
+
+    // Initializer specifically for threaded loading
+    CodeRegionEntry(uint64_t Base,
+      uint64_t Size,
+      uint64_t Offset,
+      std::string const &Filename,
+      CodeObjectSerializationHeader const &DefaultHeader)
+      : Base {Base}
+      , Size {Size}
+      , Offset {Offset}
+      , Filename {Filename}
+      , EntryHeader {DefaultHeader} {
+      }
+  };
+
+  // Map type must use an interator that isn't invalidation on erase/insert
+  using CodeRegionMapType = std::map<uint64_t, std::unique_ptr<CodeRegionEntry>>;
+  using CodeRegionPtrMapType = std::map<uint64_t, CodeRegionEntry*>;
+
+  class NamedRegionObjectHandler;
+  class CodeObjectSerializeService;
+
+  class AsyncJobHandler final {
+    public:
+      /**
+       * @brief Structure containing all the data required to async serialize code objects
+       */
+      struct SerializationJobData {
+        uint64_t GuestRIP;        ///< The RIP for the guest
+        // XXX: Support multiblock
+        uint64_t GuestCodeLength; ///< The Guest's code length
+        uint64_t GuestCodeHash;   ///< Hash of the guest code
+
+        void *HostCodeBegin;      ///< Host JIT code starting memory address
+        size_t HostCodeLength;    ///< Host JIT code length
+        uint64_t HostCodeHash;    ///< Host JIT code hash before any backpatching
+
+        // This is the thread specific ref counter for outstanding jobs.
+        // This shared mutex is incremented when the job is added, then decremented when the job is complete.
+        // If a thread is shutting down or clearing code cache then the thread will pull a unique lock on this mutex.
+        // This way it will wait until the async job handler is complete with it.
+        CodeSerializationMutex *ThreadJobRefCount;
+
+        // These are the reolocations for this serialization job
+        // Relatively small number of entries most of the time
+        std::vector<FEXCore::CPU::Relocation> Relocations;
+
+        /**
+         * @name Objects filled in from the Code Object Serialization service when a job is added
+         * @{ */
+          // This is the code region's ref counter for outstanding jobs.
+          // This shared mutex is incremented when the job is added, then decremented when the job is complete.
+          // If a named region is being removed then a unique lock will be pulled to wait for all jobs to complete and no new jobs to be added.
+          CodeSerializationMutex *ObjectJobRefCountMutexPtr;
+
+          // This is the code region iterator to reduce the number of map lookups
+          // This will remain valid while jobs are outstanding for this region
+          CodeRegionMapType::iterator CodeRegionIterator;
+        /**  @} */
+      };
+
+      AsyncJobHandler(NamedRegionObjectHandler *NamedRegionHandler, CodeObjectSerializeService *CodeObjectCacheService)
+        : NamedRegionHandler {NamedRegionHandler}
+        , CodeObjectCacheService {CodeObjectCacheService} {}
+
+    protected:
+      friend class CodeObjectSerializeService;
+      friend class NamedRegionObjectHandler;
+      /**
+       * @name Async job submission functions
+       * @{ */
+        void AsyncAddNamedRegionJob(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename);
+        void AsyncRemoveNamedRegionJob(uintptr_t Base, uintptr_t Size);
+        void AsyncAddSerializationJob(std::unique_ptr<SerializationJobData> Data);
+      /**  @} */
+
+      /**
+       * @name Async named region handling
+       * @{ */
+        /**
+         * @brief The async named region jobs to handle.
+         *
+         * Only two, Code serialization goes in to a different queue.
+         */
+        enum class NamedRegionJobType {
+          JOB_ADD_NAMED_REGION,
+          JOB_REMOVE_NAMED_REGION,
+        };
+
+        class NamedRegionWorkItem {
+          public:
+          NamedRegionJobType GetType() const { return Type; }
+
+          protected:
+            friend class WorkItemAddNamedRegion;
+            NamedRegionWorkItem(NamedRegionJobType type)
+              : Type {type} {}
+
+          private:
+            NamedRegionJobType Type;
+        };
+
+        class WorkItemAddNamedRegion : public NamedRegionWorkItem {
+          public:
+            WorkItemAddNamedRegion(const std::string &base, const std::string &filename, bool executable, CodeRegionMapType::iterator entry)
+              : NamedRegionWorkItem {NamedRegionJobType::JOB_ADD_NAMED_REGION}
+              , BaseFilename {base}
+              , Filename {filename}
+              , Executable {executable}
+              , Entry {entry}
+              {}
+            const std::string BaseFilename;
+            const std::string Filename;
+            bool Executable;
+            CodeRegionMapType::iterator Entry;
+        };
+
+        class WorkItemRemoveNamedRegion : public NamedRegionWorkItem {
+          public:
+            WorkItemRemoveNamedRegion(uint64_t base, uint64_t size, std::unique_ptr<CodeRegionEntry> entry)
+              : NamedRegionWorkItem {NamedRegionJobType::JOB_REMOVE_NAMED_REGION}
+              , Base {base}
+              , Size {size}
+              , Entry {std::move(entry)} {}
+
+            uint64_t Base;
+            uint64_t Size;
+            std::unique_ptr<CodeRegionEntry> Entry;
+        };
+      /**  @} */
+
+    private:
+      NamedRegionObjectHandler *NamedRegionHandler;
+      CodeObjectSerializeService *CodeObjectCacheService;
+  };
+
+  class NamedRegionObjectHandler final {
+    public:
+      NamedRegionObjectHandler(FEXCore::Context::Context *ctx);
+
+      void HandleNamedRegionObjectJobs();
+
+      CodeObjectSerializationConfig const &GetDefaultSerializationConfig() const {
+        return DefaultSerializationConfig;
+      }
+
+    protected:
+      friend class AsyncJobHandler;
+
+      // Return a default code header based off the default serialization config
+      CodeObjectSerializationHeader DefaultCodeHeader(uint64_t Base, uint64_t Offset) const {
+        return CodeObjectSerializationHeader {
+          .Config = DefaultSerializationConfig,
+          .OriginalBase = Base,
+          .OriginalOffset = Offset,
+          .NumCodeEntries = 0,
+          .NumRelocationsTo = 0,
+          .TotalRelocationsCount = 0,
+        };
+      }
+
+      /**
+       * @brief Adds an asynchronous add named region work item to the object queue
+       *
+       * This adds the job that will do the loading of file resources and data tracking.
+       */
+      void AsyncAddNamedRegionWorkItem(const std::string &base, const std::string &filename, bool executable, CodeRegionMapType::iterator entry) {
+        std::unique_lock lk {NamedWorkQueueMutex};
+        WorkQueue.emplace(std::make_unique<AsyncJobHandler::WorkItemAddNamedRegion> (
+          base,
+          filename,
+          executable,
+          entry
+        ));
+        ++NamedWorkQueueJobs;
+      }
+
+      void AsyncRemoveNamedRegionWorkItem(uint64_t Base, uint64_t Size, std::unique_ptr<CodeRegionEntry> Entry) {
+        std::unique_lock lk {NamedWorkQueueMutex};
+        WorkQueue.emplace(std::make_unique<AsyncJobHandler::WorkItemRemoveNamedRegion> (
+          Base,
+          Size,
+          std::move(Entry)
+        ));
+        ++NamedWorkQueueJobs;
+      }
+
+    private:
+      // Code version. If the code emission changes then this needs to increment
+      constexpr static uint32_t CODE_VERSION = 0x0;
+
+      // Default cookie header for the file header
+      constexpr static uint64_t CODE_COOKIE = FEXCore::IR::COOKIE_VERSION("FEXC", CODE_VERSION);
+
+      // Code serialization config for our current process configuration
+      CodeObjectSerializationConfig DefaultSerializationConfig;
+
+      // Atomic counter for number of jobs in the queue without needing to pull the mutex to check
+      std::atomic<uint64_t> NamedWorkQueueJobs{};
+
+      // Mutex for ading new jobs to the work queue
+      std::mutex NamedWorkQueueMutex{};
+
+      // The job queue itself
+      // Jobs get consumed as a FIFO
+      // Jobs always get appended to the end
+      std::queue<std::unique_ptr<AsyncJobHandler::NamedRegionWorkItem>> WorkQueue{};
+
+      /**
+       * @name Named Region object handling
+       * @{ */
+        void AddNamedRegionObject(CodeRegionMapType::iterator Entry, const std::string &base_filename, const std::string &filename, bool Executable);
+        void RemoveNamedRegionObject(uintptr_t Base, uintptr_t Size, std::unique_ptr<CodeRegionEntry> Entry);
+      /**  @} */
+  };
+
+  /**
+   * @brief Context specific code object serialization class
+   *
+   * Contains everything required for FEXCore to serialize code objects
+   */
+  class CodeObjectSerializeService final {
+    public:
+      CodeObjectSerializeService(FEXCore::Context::Context *ctx);
+
+      /**
+       * @brief Initialize the internal interface
+       *
+       * Is a public interface to allow the service to reinitialize after forking
+       */
+      void Initialize();
+
+      /**
+       * @brief Safely shut down the Code Object serialization service.
+       *
+       * This service needs to be resiliant to application crashes, but shutting down safely is still preferred.
+       */
+      void Shutdown();
+
+      /**
+       * @name Async interface
+       * @{ */
+        /**
+         * @brief Loads a named region in to the code serialization service. As async as possible.
+         *
+         * @param Base - Virtual address that this named region is loaded
+         * @param Size - The size of the region
+         * @param Offset - The offset from the file
+         * @param filename - The filename itself
+         */
+        void AsyncAddNamedRegionJob(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename) {
+          AsyncHandler.AsyncAddNamedRegionJob(Base, Size, Offset, filename);
+        }
+
+        /**
+         * @brief Unloads a named region from the code serialization service. As async as possible.
+         *
+         * @param Base - Virtual address of the named region
+         * @param Size - The size of the region
+         */
+        void AsyncRemoveNamedRegionJob(uintptr_t Base, uintptr_t Size) {
+          AsyncHandler.AsyncRemoveNamedRegionJob(Base, Size);
+        }
+
+        /**
+         * @brief Adds a code object serialization job. As async as possible.
+         * Code hashing happens prior to async job serialization to catch invalidations due to backpatching.
+         *
+         * @param Data - A fully filled out struct containing all the code serialization
+         */
+        void AsyncAddSerializationJob(std::unique_ptr<AsyncJobHandler::SerializationJobData> Data) {
+          AsyncHandler.AsyncAddSerializationJob(std::move(Data));
+        }
+      /**  @} */
+
+      /**
+       * @name Synchronous interface
+       * @{ */
+        /**
+         * @brief Synchronously waits for this thread's job queue to become empty.
+         *
+         * This is necessary for when a thread is shutting down
+         *
+         * @param ThreadJobRefCount - The shared mutex to wait on until to be empty
+         */
+        static void WaitForEmptyJobQueue(CodeSerializationMutex *ThreadJobRefCount) {
+          // Once the shared mutex is empty this unique lock will be gained
+          std::unique_lock lk {*ThreadJobRefCount};
+        }
+
+        /**
+         * @brief Fetches object code from the Code Object Cache for JIT.
+         *
+         * @param GuestRIP - Which GuestRIP to search the cache for
+         *
+         * @return Data required for the JIT to relocate the Object code.
+         */
+        CodeObjectFileSection const *FetchCodeObjectFromCache(uint64_t GuestRIP);
+      /**  @} */
+
+      // Public for threading
+      void ExecutionThread();
+
+    protected:
+      friend class AsyncJobHandler;
+
+      /**
+       * @brief Safely closes out code object regions from the map
+       *
+       * @param it - iterator to do a closure on
+       */
+      void DoCodeRegionClosure(uint64_t Base, CodeRegionEntry *it);
+
+      CodeSerializationMutex &GetEntryMapMutex() { return EntryMapMutex; }
+      CodeSerializationMutex &GetUnrelocatedEntryMapMutex() { return EntryMapMutex; }
+
+      CodeRegionMapType &GetEntryMap() { return AddressToEntryMap; }
+      CodeRegionPtrMapType &GetUnrelocatedEntryMap() { return UnrelocatedAddressToEntryMap; }
+
+      /**
+       * @brief Notify the async thread that it has work to do
+       */
+      void NotifyWork() { WorkAvailable.NotifyOne(); }
+
+    private:
+      FEXCore::Context::Context *CTX;
+
+      Event WorkAvailable{};
+      std::unique_ptr<FEXCore::Threads::Thread> WorkerThread;
+      std::atomic_bool WorkerThreadShuttingDown {false};
+      AsyncJobHandler AsyncHandler;
+      NamedRegionObjectHandler NamedRegionHandler;
+
+      // Mutex to hold when modifying the entry maps
+      CodeSerializationMutex EntryMapMutex;
+      CodeSerializationMutex UnrelocatedEntryMapMutex;
+
+      // Entry maps
+      CodeRegionMapType AddressToEntryMap;
+      CodeRegionPtrMapType UnrelocatedAddressToEntryMap;
+  };
+}

External/FEXCore/Source/Interface/Core/ObjectCache/Relocations.h

@@ -0,0 +1,78 @@
+#pragma once
+#include <FEXCore/IR/IR.h>
+
+namespace FEXCore::CPU {
+  enum class RelocationTypes : uint8_t {
+    // 8 byte literal in memory for symbol
+    // Aligned to struct RelocNamedSymbolLiteral
+    RELOC_NAMED_SYMBOL_LITERAL,
+
+    // Fixed size named thunk move
+    // 4 instruction constant generation on AArch64
+    // 64-bit mov on x86-64
+    // Aligned to struct RelocNamedThunkMove
+    RELOC_NAMED_THUNK_MOVE,
+
+    // Fixed size guest RIP move
+    // 4 instruction constant generation on AArch64
+    // 64-bit mov on x86-64
+    // Aligned to struct RelocGuestRIPMove
+    RELOC_GUEST_RIP_MOVE,
+  };
+
+  struct RelocationTypeHeader final {
+    RelocationTypes Type;
+  };
+
+  struct RelocNamedSymbolLiteral final {
+    enum class NamedSymbol : uint8_t {
+      ///< Thread specific relocations
+      // JIT Literal pointers
+      SYMBOL_LITERAL_EXITFUNCTION_LINKER,
+    };
+
+    RelocationTypeHeader Header{};
+
+    NamedSymbol Symbol;
+
+    // Offset in to the code section to begin the relocation
+    uint64_t Offset{};
+  };
+
+  struct RelocNamedThunkMove final {
+    RelocationTypeHeader Header{};
+
+    // GPR index the constant is being moved to
+    uint8_t RegisterIndex;
+
+    // The thunk SHA256 hash
+    IR::SHA256Sum Symbol;
+
+    // Offset in to the code section to begin the relocation
+    uint64_t Offset{};
+  };
+
+  struct RelocGuestRIPMove final {
+    RelocationTypeHeader Header{};
+
+    // GPR index the constant is being moved to
+    uint8_t RegisterIndex;
+
+    // Offset in to the code section to begin the relocation
+    uint64_t Offset{};
+
+    // The unrelocated RIP that is being moved
+    uint64_t GuestRIP;
+  };
+
+  union Relocation {
+    RelocationTypeHeader Header{};
+
+    RelocNamedSymbolLiteral NamedSymbolLiteral;
+    // This makes our union of relocations at least 48 bytes
+    // It might be more efficient to not use a union
+    RelocNamedThunkMove NamedThunkMove;
+
+    RelocGuestRIPMove GuestRIPMove;
+  };
+}

External/FEXCore/Source/Interface/Core/OpcodeDispatcher.cpp

@@ -1443,6 +1443,11 @@ void OpDispatchBuilder::XCHGOp(OpcodeArgs) {
     // But this would result in a zext on 64bit, which would ruin the no-op nature of the instruction
     // So x86-64 spec mandates this special case that even though it is a 32bit instruction and
     // is supposed to zext the result, it is a true no-op
+    if (Op->Flags & FEXCore::X86Tables::DecodeFlags::FLAG_REP_PREFIX) {
+      // If this instruction has a REP prefix then this is architectually defined to be a `PAUSE` instruction.
+      // On older processors this ends up being a true `REP NOP` which is why they stuck this here.
+      _Yield();
+    }
     return;
   }
 
@@ -3353,7 +3358,9 @@ void OpDispatchBuilder::ReadSegmentReg(OpcodeArgs) {
 
 template<OpDispatchBuilder::Segment Seg>
 void OpDispatchBuilder::WriteSegmentReg(OpcodeArgs) {
-  auto Size = GetSrcSize(Op);
+  // Documentation claims that the 32-bit version of this instruction inserts in to the lower 32-bits of the segment
+  // This is incorrect and it instead zero extends the 32-bit value to 64-bit
+  auto Size = GetDstSize(Op);
   OrderedNode *Src = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, -1);
   if constexpr (Seg == Segment::FS) {
     _StoreContext(Size, GPRClass, Src, offsetof(FEXCore::Core::CPUState, fs));
@@ -4604,7 +4611,7 @@ OrderedNode *OpDispatchBuilder::AppendSegmentOffset(OrderedNode *Value, uint32_t
   return Value;
 }
 
-OrderedNode *OpDispatchBuilder::LoadSource_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint8_t OpSize, uint32_t Flags, int8_t Align, bool LoadData, bool ForceLoad) {
+OrderedNode *OpDispatchBuilder::LoadSource_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint8_t OpSize, uint32_t Flags, int8_t Align, bool LoadData, bool ForceLoad, MemoryAccessType AccessType) {
   LOGMAN_THROW_A_FMT(Operand.IsGPR() ||
                      Operand.IsLiteral() ||
                      Operand.IsGPRDirect() ||
@@ -4615,7 +4622,6 @@ OrderedNode *OpDispatchBuilder::LoadSource_WithOpSize(FEXCore::IR::RegisterClass
 
   OrderedNode *Src {nullptr};
   bool LoadableType = false;
-  bool StackAccess = false;
   const uint8_t GPRSize = CTX->GetGPRSize();
   const uint32_t AddrSize = (Op->Flags & X86Tables::DecodeFlags::FLAG_ADDRESS_SIZE) != 0 ? (GPRSize >> 1) : GPRSize;
 
@@ -4644,7 +4650,9 @@ OrderedNode *OpDispatchBuilder::LoadSource_WithOpSize(FEXCore::IR::RegisterClass
   else if (Operand.IsGPRDirect()) {
     Src = _LoadContext(AddrSize, GPRClass, offsetof(FEXCore::Core::CPUState, gregs[Operand.Data.GPR.GPR]));
     LoadableType = true;
-    StackAccess = Operand.Data.GPR.GPR == FEXCore::X86State::REG_RSP;
+    if (Operand.Data.GPR.GPR == FEXCore::X86State::REG_RSP && AccessType == MemoryAccessType::ACCESS_DEFAULT) {
+      AccessType = MemoryAccessType::ACCESS_NONTSO;
+    }
   }
   else if (Operand.IsGPRIndirect()) {
     auto GPR = _LoadContext(AddrSize, GPRClass, offsetof(FEXCore::Core::CPUState, gregs[Operand.Data.GPRIndirect.GPR]));
@@ -4653,7 +4661,9 @@ OrderedNode *OpDispatchBuilder::LoadSource_WithOpSize(FEXCore::IR::RegisterClass
 		Src = _Add(GPR, Constant);
 
     LoadableType = true;
-    StackAccess = Operand.Data.GPRIndirect.GPR == FEXCore::X86State::REG_RSP;
+    if (Operand.Data.GPRIndirect.GPR == FEXCore::X86State::REG_RSP && AccessType == MemoryAccessType::ACCESS_DEFAULT) {
+      AccessType = MemoryAccessType::ACCESS_NONTSO;
+    }
   }
   else if (Operand.IsRIPRelative()) {
     if (CTX->Config.Is64BitMode) {
@@ -4675,7 +4685,9 @@ OrderedNode *OpDispatchBuilder::LoadSource_WithOpSize(FEXCore::IR::RegisterClass
         auto Constant = _Constant(GPRSize * 8, Operand.Data.SIB.Scale);
         Tmp = _Mul(Tmp, Constant);
       }
-      StackAccess |= Operand.Data.SIB.Index == FEXCore::X86State::REG_RSP;
+      if (Operand.Data.SIB.Index == FEXCore::X86State::REG_RSP && AccessType == MemoryAccessType::ACCESS_DEFAULT) {
+        AccessType = MemoryAccessType::ACCESS_NONTSO;
+      }
     }
 
     if (Operand.Data.SIB.Base != FEXCore::X86State::REG_INVALID) {
@@ -4687,7 +4699,10 @@ OrderedNode *OpDispatchBuilder::LoadSource_WithOpSize(FEXCore::IR::RegisterClass
       else {
         Tmp = GPR;
       }
-      StackAccess |= Operand.Data.SIB.Base == FEXCore::X86State::REG_RSP;
+
+      if (Operand.Data.SIB.Base == FEXCore::X86State::REG_RSP && AccessType == MemoryAccessType::ACCESS_DEFAULT) {
+        AccessType = MemoryAccessType::ACCESS_NONTSO;
+      }
     }
 
     if (Operand.Data.SIB.Offset) {
@@ -4722,7 +4737,7 @@ OrderedNode *OpDispatchBuilder::LoadSource_WithOpSize(FEXCore::IR::RegisterClass
   if ((LoadableType && LoadData) || ForceLoad) {
     Src = AppendSegmentOffset(Src, Flags);
 
-    if (StackAccess) {
+    if (AccessType == MemoryAccessType::ACCESS_NONTSO || AccessType == MemoryAccessType::ACCESS_STREAM) {
       Src = _LoadMem(Class, OpSize, Src, Align == -1 ? OpSize : Align);
     }
     else {
@@ -4737,12 +4752,12 @@ OrderedNode *OpDispatchBuilder::GetRelocatedPC(FEXCore::X86Tables::DecodedOp con
   return _EntrypointOffset(Op->PC + Op->InstSize + Offset - Entry, GPRSize);
 }
 
-OrderedNode *OpDispatchBuilder::LoadSource(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint32_t Flags, int8_t Align, bool LoadData, bool ForceLoad) {
+OrderedNode *OpDispatchBuilder::LoadSource(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint32_t Flags, int8_t Align, bool LoadData, bool ForceLoad, MemoryAccessType AccessType) {
   const uint8_t OpSize = GetSrcSize(Op);
-  return LoadSource_WithOpSize(Class, Op, Operand, OpSize, Flags, Align, LoadData, ForceLoad);
+  return LoadSource_WithOpSize(Class, Op, Operand, OpSize, Flags, Align, LoadData, ForceLoad, AccessType);
 }
 
-void OpDispatchBuilder::StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, FEXCore::X86Tables::DecodedOperand const& Operand, OrderedNode *const Src, uint8_t OpSize, int8_t Align) {
+void OpDispatchBuilder::StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, FEXCore::X86Tables::DecodedOperand const& Operand, OrderedNode *const Src, uint8_t OpSize, int8_t Align, MemoryAccessType AccessType) {
   LOGMAN_THROW_A_FMT(Operand.IsGPR() ||
                      Operand.IsLiteral() ||
                      Operand.IsGPRDirect() ||
@@ -4755,7 +4770,6 @@ void OpDispatchBuilder::StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Cl
   // 32bit ops ZEXT the result to 64bit
   OrderedNode *MemStoreDst {nullptr};
   bool MemStore = false;
-  bool StackAccess = false;
   const uint8_t GPRSize = CTX->GetGPRSize();
   const uint32_t AddrSize = (Op->Flags & X86Tables::DecodeFlags::FLAG_ADDRESS_SIZE) != 0 ? (GPRSize >> 1) : GPRSize;
 
@@ -4789,7 +4803,9 @@ void OpDispatchBuilder::StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Cl
   else if (Operand.IsGPRDirect()) {
     MemStoreDst = _LoadContext(AddrSize, GPRClass, offsetof(FEXCore::Core::CPUState, gregs[Operand.Data.GPR.GPR]));
     MemStore = true;
-    StackAccess = Operand.Data.GPR.GPR == FEXCore::X86State::REG_RSP;
+    if (Operand.Data.GPR.GPR == FEXCore::X86State::REG_RSP && AccessType == MemoryAccessType::ACCESS_DEFAULT) {
+      AccessType = MemoryAccessType::ACCESS_NONTSO;
+    }
   }
   else if (Operand.IsGPRIndirect()) {
     auto GPR = _LoadContext(AddrSize, GPRClass, offsetof(FEXCore::Core::CPUState, gregs[Operand.Data.GPRIndirect.GPR]));
@@ -4797,7 +4813,9 @@ void OpDispatchBuilder::StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Cl
 
     MemStoreDst = _Add(GPR, Constant);
     MemStore = true;
-    StackAccess = Operand.Data.GPRIndirect.GPR == FEXCore::X86State::REG_RSP;
+    if (Operand.Data.GPRIndirect.GPR == FEXCore::X86State::REG_RSP && AccessType == MemoryAccessType::ACCESS_DEFAULT) {
+      AccessType = MemoryAccessType::ACCESS_NONTSO;
+    }
   }
   else if (Operand.IsRIPRelative()) {
     if (CTX->Config.Is64BitMode) {
@@ -4867,7 +4885,7 @@ void OpDispatchBuilder::StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Cl
       auto DestAddr = _Add(MemStoreDst, _Constant(8));
       _StoreMem(GPRClass, 2, DestAddr, Upper, std::min<uint8_t>(Align, 8));
     } else {
-      if (StackAccess) {
+      if (AccessType == MemoryAccessType::ACCESS_NONTSO || AccessType == MemoryAccessType::ACCESS_STREAM) {
         _StoreMem(Class, OpSize, MemStoreDst, Src, Align == -1 ? OpSize : Align);
       }
       else {
@@ -4877,17 +4895,23 @@ void OpDispatchBuilder::StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Cl
   }
 }
 
-void OpDispatchBuilder::StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, FEXCore::X86Tables::DecodedOperand const& Operand, OrderedNode *const Src, int8_t Align) {
-  StoreResult_WithOpSize(Class, Op, Operand, Src, GetDstSize(Op), Align);
+void OpDispatchBuilder::StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, FEXCore::X86Tables::DecodedOperand const& Operand, OrderedNode *const Src, int8_t Align, MemoryAccessType AccessType) {
+  StoreResult_WithOpSize(Class, Op, Operand, Src, GetDstSize(Op), Align, AccessType);
 }
 
-void OpDispatchBuilder::StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, OrderedNode *const Src, int8_t Align) {
-  StoreResult(Class, Op, Op->Dest, Src, Align);
+void OpDispatchBuilder::StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, OrderedNode *const Src, int8_t Align, MemoryAccessType AccessType) {
+  StoreResult(Class, Op, Op->Dest, Src, Align, AccessType);
 }
 
 OpDispatchBuilder::OpDispatchBuilder(FEXCore::Context::Context *ctx)
-  : CTX {ctx} {
+  : IREmitter {ctx->OpDispatcherAllocator}
+  , CTX {ctx} {
   ResetWorkingList();
+  InstallHostSpecificOpcodeHandlers();
+}
+OpDispatchBuilder::OpDispatchBuilder(FEXCore::Utils::IntrusivePooledAllocator &Allocator)
+  : IREmitter {Allocator}
+  , CTX {nullptr} {
 }
 
 void OpDispatchBuilder::ResetWorkingList() {
@@ -4909,6 +4933,11 @@ void OpDispatchBuilder::MOVGPROp(OpcodeArgs) {
   StoreResult(GPRClass, Op, Src, 1);
 }
 
+void OpDispatchBuilder::MOVGPRNTOp(OpcodeArgs) {
+  OrderedNode *Src = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, 1);
+  StoreResult(GPRClass, Op, Src, 1, MemoryAccessType::ACCESS_STREAM);
+}
+
 void OpDispatchBuilder::ALUOp(OpcodeArgs) {
   bool RequiresMask = false;
   FEXCore::IR::IROps IROp;
@@ -5233,6 +5262,103 @@ void OpDispatchBuilder::InvalidOp(OpcodeArgs) {
 
 #undef OpcodeArgs
 
+
+void OpDispatchBuilder::InstallHostSpecificOpcodeHandlers() {
+  static bool Initialized = false;
+  if (!CTX || Initialized) {
+    // IRCompaction doesn't set a CTX and doesn't need this anyway
+    return;
+  }
+#define OPD(prefix, opcode) (((prefix) << 8) | opcode)
+  constexpr uint16_t PF_38_NONE = 0;
+  constexpr uint16_t PF_38_66   = (1U << 0);
+  constexpr uint16_t PF_38_F2   = (1U << 1);
+
+  constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> H0F38_SHA[] = {
+    {OPD(PF_38_NONE, 0xC8), 1, &OpDispatchBuilder::SHA1NEXTEOp},
+    {OPD(PF_38_NONE, 0xC9), 1, &OpDispatchBuilder::SHA1MSG1Op},
+    {OPD(PF_38_NONE, 0xCA), 1, &OpDispatchBuilder::SHA1MSG2Op},
+    {OPD(PF_38_NONE, 0xCB), 1, &OpDispatchBuilder::SHA256RNDS2Op},
+    {OPD(PF_38_NONE, 0xCC), 1, &OpDispatchBuilder::SHA256MSG1Op},
+    {OPD(PF_38_NONE, 0xCD), 1, &OpDispatchBuilder::SHA256MSG2Op},
+  };
+
+  constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> H0F38_AES[] = {
+    {OPD(PF_38_66, 0xDB), 1, &OpDispatchBuilder::AESImcOp},
+    {OPD(PF_38_66, 0xDC), 1, &OpDispatchBuilder::AESEncOp},
+    {OPD(PF_38_66, 0xDD), 1, &OpDispatchBuilder::AESEncLastOp},
+    {OPD(PF_38_66, 0xDE), 1, &OpDispatchBuilder::AESDecOp},
+    {OPD(PF_38_66, 0xDF), 1, &OpDispatchBuilder::AESDecLastOp},
+  };
+  constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> H0F38_CRC[] = {
+    {OPD(PF_38_F2, 0xF0), 1, &OpDispatchBuilder::CRC32},
+    {OPD(PF_38_F2, 0xF1), 1, &OpDispatchBuilder::CRC32},
+
+    {OPD(PF_38_66 | PF_38_F2, 0xF0), 1, &OpDispatchBuilder::CRC32},
+    {OPD(PF_38_66 | PF_38_F2, 0xF1), 1, &OpDispatchBuilder::CRC32},
+  };
+#undef OPD
+
+#define OPD(REX, prefix, opcode) ((REX << 9) | (prefix << 8) | opcode)
+#define PF_3A_NONE 0
+#define PF_3A_66   1
+  constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> H0F3A_AES[] = {
+    {OPD(0, PF_3A_66,   0xDF), 1, &OpDispatchBuilder::AESKeyGenAssist},
+  };
+#undef PF_3A_NONE
+#undef PF_3A_66
+#undef OPD
+
+#define OPD(group, prefix, Reg) (((group - FEXCore::X86Tables::TYPE_GROUP_6) << 5) | (prefix) << 3 | (Reg))
+  constexpr uint16_t PF_NONE = 0;
+  constexpr uint16_t PF_66 = 2;
+
+  constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> SecondaryExtensionOp_RDRAND[] = {
+    // GROUP 9
+    {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_NONE, 6), 1, &OpDispatchBuilder::RDRANDOp<false>},
+    {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_NONE, 7), 1, &OpDispatchBuilder::RDRANDOp<true>},
+
+    {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_66, 6), 1, &OpDispatchBuilder::RDRANDOp<false>},
+    {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_66, 7), 1, &OpDispatchBuilder::RDRANDOp<true>},
+  };
+#undef OPD
+
+  constexpr std::tuple<uint8_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> SecondaryModRMExtensionOp_CLZero[] = {
+    {((3 << 3) | 4), 1, &OpDispatchBuilder::CLZeroOp},
+  };
+
+  auto InstallToTable = [](auto& FinalTable, auto& LocalTable) {
+    for (auto Op : LocalTable) {
+      auto OpNum = std::get<0>(Op);
+      auto Dispatcher = std::get<2>(Op);
+      for (uint8_t i = 0; i < std::get<1>(Op); ++i) {
+        LOGMAN_THROW_A_FMT(FinalTable[OpNum + i].OpcodeDispatcher == nullptr, "Duplicate Entry");
+        FinalTable[OpNum + i].OpcodeDispatcher = Dispatcher;
+      }
+    }
+  };
+
+  if (CTX->HostFeatures.SupportsCRC) {
+    InstallToTable(FEXCore::X86Tables::H0F38TableOps, H0F38_CRC);
+  }
+
+  InstallToTable(FEXCore::X86Tables::H0F38TableOps, H0F38_SHA);
+
+  if (CTX->HostFeatures.SupportsAES) {
+    InstallToTable(FEXCore::X86Tables::H0F38TableOps, H0F38_AES);
+    InstallToTable(FEXCore::X86Tables::H0F3ATableOps, H0F3A_AES);
+  }
+
+  if (CTX->HostFeatures.SupportsCLZERO) {
+    InstallToTable(FEXCore::X86Tables::SecondModRMTableOps, SecondaryModRMExtensionOp_CLZero);
+  }
+
+  if (CTX->HostFeatures.SupportsRAND) {
+    InstallToTable(FEXCore::X86Tables::SecondInstGroupOps, SecondaryExtensionOp_RDRAND);
+  }
+  Initialized = true;
+}
+
 void InstallOpcodeHandlers(Context::OperatingMode Mode) {
   constexpr std::tuple<uint8_t, uint8_t, X86Tables::OpDispatchPtr> BaseOpTable[] = {
     // Instructions
@@ -5361,7 +5487,7 @@ void InstallOpcodeHandlers(Context::OperatingMode Mode) {
     {0xBD, 1, &OpDispatchBuilder::BSROp}, // BSF
     {0xBE, 2, &OpDispatchBuilder::MOVSXOp},
     {0xC0, 2, &OpDispatchBuilder::XADDOp},
-    {0xC3, 1, &OpDispatchBuilder::MOVGPROp<0>},
+    {0xC3, 1, &OpDispatchBuilder::MOVGPRNTOp},
     {0xC4, 1, &OpDispatchBuilder::PINSROp<2>},
     {0xC5, 1, &OpDispatchBuilder::PExtrOp<2>},
     {0xC8, 8, &OpDispatchBuilder::BSWAPOp},
@@ -5375,7 +5501,7 @@ void InstallOpcodeHandlers(Context::OperatingMode Mode) {
     {0x17, 1, &OpDispatchBuilder::MOVUPSOp},
     {0x28, 2, &OpDispatchBuilder::MOVUPSOp},
     {0x2A, 1, &OpDispatchBuilder::MMX_To_XMM_Vector_CVT_Int_To_Float<4, false>},
-    {0x2B, 1, &OpDispatchBuilder::MOVAPSOp},
+    {0x2B, 1, &OpDispatchBuilder::MOVVectorNTOp},
     {0x2C, 1, &OpDispatchBuilder::Vector_CVT_Float_To_Int<4, false, false>},
     {0x2D, 1, &OpDispatchBuilder::Vector_CVT_Float_To_Int<4, false, true>},
     {0x2E, 2, &OpDispatchBuilder::UCOMISxOp<4>},
@@ -5437,7 +5563,7 @@ void InstallOpcodeHandlers(Context::OperatingMode Mode) {
     {0xE3, 1, &OpDispatchBuilder::PAVGOp<2>},
     {0xE4, 1, &OpDispatchBuilder::PMULHW<false>},
     {0xE5, 1, &OpDispatchBuilder::PMULHW<true>},
-    {0xE7, 1, &OpDispatchBuilder::MOVUPSOp},
+    {0xE7, 1, &OpDispatchBuilder::MOVVectorNTOp},
     {0xE8, 1, &OpDispatchBuilder::PSUBSOp<1, true>},
     {0xE9, 1, &OpDispatchBuilder::PSUBSOp<2, true>},
     {0xEA, 1, &OpDispatchBuilder::VectorALUOp<IR::OP_VSMIN, 2>},
@@ -5665,7 +5791,7 @@ void InstallOpcodeHandlers(Context::OperatingMode Mode) {
     {0x19, 7, &OpDispatchBuilder::NOPOp},
     {0x28, 2, &OpDispatchBuilder::MOVAPSOp},
     {0x2A, 1, &OpDispatchBuilder::MMX_To_XMM_Vector_CVT_Int_To_Float<4, true>},
-    {0x2B, 1, &OpDispatchBuilder::MOVAPSOp},
+    {0x2B, 1, &OpDispatchBuilder::MOVVectorNTOp},
     {0x2C, 1, &OpDispatchBuilder::XMM_To_MMX_Vector_CVT_Float_To_Int<8, false>},
     {0x2D, 1, &OpDispatchBuilder::XMM_To_MMX_Vector_CVT_Float_To_Int<8, true>},
     {0x2E, 2, &OpDispatchBuilder::UCOMISxOp<8>},
@@ -5739,7 +5865,7 @@ void InstallOpcodeHandlers(Context::OperatingMode Mode) {
     {0xE4, 1, &OpDispatchBuilder::PMULHW<false>},
     {0xE5, 1, &OpDispatchBuilder::PMULHW<true>},
     {0xE6, 1, &OpDispatchBuilder::Vector_CVT_Float_To_Int<8, true, false>},
-    {0xE7, 1, &OpDispatchBuilder::MOVVectorOp},
+    {0xE7, 1, &OpDispatchBuilder::MOVVectorNTOp},
     {0xE8, 1, &OpDispatchBuilder::VectorALUOp<IR::OP_VSQSUB, 1>},
     {0xE9, 1, &OpDispatchBuilder::VectorALUOp<IR::OP_VSQSUB, 2>},
     {0xEA, 1, &OpDispatchBuilder::VectorALUOp<IR::OP_VSMIN, 2>},
@@ -5794,15 +5920,8 @@ constexpr uint16_t PF_F2 = 3;
 
     // GROUP 9
     {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_NONE, 1), 1, &OpDispatchBuilder::CMPXCHGPairOp},
-    {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_NONE, 6), 1, &OpDispatchBuilder::RDRANDOp<false>},
-    {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_NONE, 7), 1, &OpDispatchBuilder::RDRANDOp<true>},
-
     {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_F3, 1), 1, &OpDispatchBuilder::CMPXCHGPairOp},
-
     {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_66, 1), 1, &OpDispatchBuilder::CMPXCHGPairOp},
-    {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_66, 6), 1, &OpDispatchBuilder::RDRANDOp<false>},
-    {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_66, 7), 1, &OpDispatchBuilder::RDRANDOp<true>},
-
     {OPD(FEXCore::X86Tables::TYPE_GROUP_9, PF_F2, 1), 1, &OpDispatchBuilder::CMPXCHGPairOp},
 
     // GROUP 12
@@ -5841,10 +5960,6 @@ constexpr uint16_t PF_F2 = 3;
     {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_NONE, 6), 1, &OpDispatchBuilder::FenceOp<FEXCore::IR::Fence_LoadStore.Val>}, //MFENCE
     {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_NONE, 7), 1, &OpDispatchBuilder::StoreFenceOrCLFlush},     //SFENCE
 
-    {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 0), 1, &OpDispatchBuilder::ReadSegmentReg<OpDispatchBuilder::Segment::FS>},
-    {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 1), 1, &OpDispatchBuilder::ReadSegmentReg<OpDispatchBuilder::Segment::GS>},
-    {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 2), 1, &OpDispatchBuilder::WriteSegmentReg<OpDispatchBuilder::Segment::FS>},
-    {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 3), 1, &OpDispatchBuilder::WriteSegmentReg<OpDispatchBuilder::Segment::GS>},
     {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 5), 1, &OpDispatchBuilder::UnimplementedOp},
     {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 6), 1, &OpDispatchBuilder::UnimplementedOp},
 
@@ -5860,6 +5975,15 @@ constexpr uint16_t PF_F2 = 3;
     {OPD(FEXCore::X86Tables::TYPE_GROUP_P, PF_66, 0), 8, &OpDispatchBuilder::NOPOp},
     {OPD(FEXCore::X86Tables::TYPE_GROUP_P, PF_F2, 0), 8, &OpDispatchBuilder::NOPOp},
   };
+
+  constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> SecondaryExtensionOpTable_64[] = {
+    // GROUP 15
+    {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 0), 1, &OpDispatchBuilder::ReadSegmentReg<OpDispatchBuilder::Segment::FS>},
+    {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 1), 1, &OpDispatchBuilder::ReadSegmentReg<OpDispatchBuilder::Segment::GS>},
+    {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 2), 1, &OpDispatchBuilder::WriteSegmentReg<OpDispatchBuilder::Segment::FS>},
+    {OPD(FEXCore::X86Tables::TYPE_GROUP_15, PF_F3, 3), 1, &OpDispatchBuilder::WriteSegmentReg<OpDispatchBuilder::Segment::GS>},
+  };
+
 #undef OPD
 
   constexpr std::tuple<uint8_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> SecondaryModRMExtensionOpTable[] = {
@@ -5868,13 +5992,242 @@ constexpr uint16_t PF_F2 = 3;
 
     // REG /7
     {((3 << 3) | 1), 1, &OpDispatchBuilder::RDTSCPOp},
-    {((3 << 3) | 4), 1, &OpDispatchBuilder::CLZeroOp},
 
   };
 // Top bit indicating if it needs to be repeated with {0x40, 0x80} or'd in
 // All OPDReg versions need it
 #define OPDReg(op, reg) ((1 << 15) | ((op - 0xD8) << 8) | (reg << 3))
 #define OPD(op, modrmop) (((op - 0xD8) << 8) | modrmop)
+  constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> X87F64OpTable[] = {
+    {OPDReg(0xD8, 0) | 0x00, 8, &OpDispatchBuilder::FADDF64<32, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xD8, 1) | 0x00, 8, &OpDispatchBuilder::FMULF64<32, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xD8, 2) | 0x00, 8, &OpDispatchBuilder::FCOMIF64<32, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xD8, 3) | 0x00, 8, &OpDispatchBuilder::FCOMIF64<32, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xD8, 4) | 0x00, 8, &OpDispatchBuilder::FSUBF64<32, false, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xD8, 5) | 0x00, 8, &OpDispatchBuilder::FSUBF64<32, false, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xD8, 6) | 0x00, 8, &OpDispatchBuilder::FDIVF64<32, false, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xD8, 7) | 0x00, 8, &OpDispatchBuilder::FDIVF64<32, false, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+      {OPD(0xD8, 0xC0), 8, &OpDispatchBuilder::FADDF64<80, false, OpDispatchBuilder::OpResult::RES_ST0>},
+      {OPD(0xD8, 0xC8), 8, &OpDispatchBuilder::FMULF64<80, false, OpDispatchBuilder::OpResult::RES_ST0>},
+      {OPD(0xD8, 0xD0), 8, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+      {OPD(0xD8, 0xD8), 8, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+      {OPD(0xD8, 0xE0), 8, &OpDispatchBuilder::FSUBF64<80, false, false, OpDispatchBuilder::OpResult::RES_ST0>},
+      {OPD(0xD8, 0xE8), 8, &OpDispatchBuilder::FSUBF64<80, false, true, OpDispatchBuilder::OpResult::RES_ST0>},
+      {OPD(0xD8, 0xF0), 8, &OpDispatchBuilder::FDIVF64<80, false, false, OpDispatchBuilder::OpResult::RES_ST0>},
+      {OPD(0xD8, 0xF8), 8, &OpDispatchBuilder::FDIVF64<80, false, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xD9, 0) | 0x00, 8, &OpDispatchBuilder::FLDF64<32>},
+
+    // 1 = Invalid
+
+    {OPDReg(0xD9, 2) | 0x00, 8, &OpDispatchBuilder::FSTF64<32>},
+
+    {OPDReg(0xD9, 3) | 0x00, 8, &OpDispatchBuilder::FSTF64<32>},
+
+    {OPDReg(0xD9, 4) | 0x00, 8, &OpDispatchBuilder::X87LDENVF64},
+
+    {OPDReg(0xD9, 5) | 0x00, 8, &OpDispatchBuilder::X87FLDCWF64},
+
+    {OPDReg(0xD9, 6) | 0x00, 8, &OpDispatchBuilder::X87FNSTENV},
+
+    {OPDReg(0xD9, 7) | 0x00, 8, &OpDispatchBuilder::X87FSTCW},
+
+      {OPD(0xD9, 0xC0), 8, &OpDispatchBuilder::FLDF64<80>},
+      {OPD(0xD9, 0xC8), 8, &OpDispatchBuilder::FXCH},
+      {OPD(0xD9, 0xD0), 1, &OpDispatchBuilder::NOPOp}, // FNOP
+      // D1 = Invalid
+      // D8 = Invalid
+      {OPD(0xD9, 0xE0), 1, &OpDispatchBuilder::FCHSF64},
+      {OPD(0xD9, 0xE1), 1, &OpDispatchBuilder::FABSF64},
+      // E2 = Invalid
+      {OPD(0xD9, 0xE4), 1, &OpDispatchBuilder::FTSTF64},
+      {OPD(0xD9, 0xE5), 1, &OpDispatchBuilder::X87FXAMF64},
+      // E6 = Invalid
+      {OPD(0xD9, 0xE8), 1, &OpDispatchBuilder::FLDF64_Const<0x3FF0000000000000>}, // 1.0
+      {OPD(0xD9, 0xE9), 1, &OpDispatchBuilder::FLDF64_Const<0x400A934F0979A372>}, // log2l(10)
+      {OPD(0xD9, 0xEA), 1, &OpDispatchBuilder::FLDF64_Const<0x3FF71547652B82FE>}, // log2l(e)
+      {OPD(0xD9, 0xEB), 1, &OpDispatchBuilder::FLDF64_Const<0x400921FB54442D18>}, // pi
+      {OPD(0xD9, 0xEC), 1, &OpDispatchBuilder::FLDF64_Const<0x3FD34413509F79FF>}, // log10l(2)
+      {OPD(0xD9, 0xED), 1, &OpDispatchBuilder::FLDF64_Const<0x3FE62E42FEFA39EF>}, // log(2)
+      {OPD(0xD9, 0xEE), 1, &OpDispatchBuilder::FLDF64_Const<0>}, // 0.0
+
+      // EF = Invalid
+      {OPD(0xD9, 0xF0), 1, &OpDispatchBuilder::X87UnaryOpF64<IR::OP_F64F2XM1>},
+      {OPD(0xD9, 0xF1), 1, &OpDispatchBuilder::X87FYL2XF64},
+      {OPD(0xD9, 0xF2), 1, &OpDispatchBuilder::X87TANF64},
+      {OPD(0xD9, 0xF3), 1, &OpDispatchBuilder::X87ATANF64},
+      {OPD(0xD9, 0xF4), 1, &OpDispatchBuilder::FXTRACTF64},
+      {OPD(0xD9, 0xF5), 1, &OpDispatchBuilder::X87BinaryOpF64<IR::OP_F64FPREM1>},
+      {OPD(0xD9, 0xF6), 1, &OpDispatchBuilder::X87ModifySTP<false>},
+      {OPD(0xD9, 0xF7), 1, &OpDispatchBuilder::X87ModifySTP<true>},
+      {OPD(0xD9, 0xF8), 1, &OpDispatchBuilder::X87BinaryOpF64<IR::OP_F64FPREM>},
+      {OPD(0xD9, 0xF9), 1, &OpDispatchBuilder::X87FYL2XF64},
+      {OPD(0xD9, 0xFA), 1, &OpDispatchBuilder::FSQRTF64},
+      {OPD(0xD9, 0xFB), 1, &OpDispatchBuilder::X87SinCosF64},
+      {OPD(0xD9, 0xFC), 1, &OpDispatchBuilder::FRNDINTF64},
+      {OPD(0xD9, 0xFD), 1, &OpDispatchBuilder::X87BinaryOpF64<IR::OP_F64SCALE>},
+      {OPD(0xD9, 0xFE), 1, &OpDispatchBuilder::X87UnaryOpF64<IR::OP_F64SIN>},
+      {OPD(0xD9, 0xFF), 1, &OpDispatchBuilder::X87UnaryOpF64<IR::OP_F64COS>},
+
+    {OPDReg(0xDA, 0) | 0x00, 8, &OpDispatchBuilder::FADDF64<32, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDA, 1) | 0x00, 8, &OpDispatchBuilder::FMULF64<32, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDA, 2) | 0x00, 8, &OpDispatchBuilder::FCOMIF64<32, true, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xDA, 3) | 0x00, 8, &OpDispatchBuilder::FCOMIF64<32, true, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xDA, 4) | 0x00, 8, &OpDispatchBuilder::FSUBF64<32, true, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDA, 5) | 0x00, 8, &OpDispatchBuilder::FSUBF64<32, true, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDA, 6) | 0x00, 8, &OpDispatchBuilder::FDIVF64<32, true, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDA, 7) | 0x00, 8, &OpDispatchBuilder::FDIVF64<32, true, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+      {OPD(0xDA, 0xC0), 8, &OpDispatchBuilder::X87FCMOV},
+      {OPD(0xDA, 0xC8), 8, &OpDispatchBuilder::X87FCMOV},
+      {OPD(0xDA, 0xD0), 8, &OpDispatchBuilder::X87FCMOV},
+      {OPD(0xDA, 0xD8), 8, &OpDispatchBuilder::X87FCMOV},
+      // E0 = Invalid
+      // E8 = Invalid
+      {OPD(0xDA, 0xE9), 1, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, true>},
+      // EA = Invalid
+      // F0 = Invalid
+      // F8 = Invalid
+
+    {OPDReg(0xDB, 0) | 0x00, 8, &OpDispatchBuilder::FILDF64},
+
+    {OPDReg(0xDB, 1) | 0x00, 8, &OpDispatchBuilder::FISTF64<true>},
+
+    {OPDReg(0xDB, 2) | 0x00, 8, &OpDispatchBuilder::FISTF64<false>},
+
+    {OPDReg(0xDB, 3) | 0x00, 8, &OpDispatchBuilder::FISTF64<false>},
+
+    // 4 = Invalid
+
+    {OPDReg(0xDB, 5) | 0x00, 8, &OpDispatchBuilder::FLDF64<80>},
+
+    // 6 = Invalid
+
+    {OPDReg(0xDB, 7) | 0x00, 8, &OpDispatchBuilder::FSTF64<80>},
+
+
+      {OPD(0xDB, 0xC0), 8, &OpDispatchBuilder::X87FCMOV},
+      {OPD(0xDB, 0xC8), 8, &OpDispatchBuilder::X87FCMOV},
+      {OPD(0xDB, 0xD0), 8, &OpDispatchBuilder::X87FCMOV},
+      {OPD(0xDB, 0xD8), 8, &OpDispatchBuilder::X87FCMOV},
+      // E0 = Invalid
+      {OPD(0xDB, 0xE2), 1, &OpDispatchBuilder::NOPOp}, // FNCLEX
+      {OPD(0xDB, 0xE3), 1, &OpDispatchBuilder::FNINITF64},
+      // E4 = Invalid
+      {OPD(0xDB, 0xE8), 8, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_RFLAGS, false>},
+      {OPD(0xDB, 0xF0), 8, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_RFLAGS, false>},
+
+      // F8 = Invalid
+
+    {OPDReg(0xDC, 0) | 0x00, 8, &OpDispatchBuilder::FADDF64<64, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDC, 1) | 0x00, 8, &OpDispatchBuilder::FMULF64<64, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDC, 2) | 0x00, 8, &OpDispatchBuilder::FCOMIF64<64, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xDC, 3) | 0x00, 8, &OpDispatchBuilder::FCOMIF64<64, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xDC, 4) | 0x00, 8, &OpDispatchBuilder::FSUBF64<64, false, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDC, 5) | 0x00, 8, &OpDispatchBuilder::FSUBF64<64, false, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDC, 6) | 0x00, 8, &OpDispatchBuilder::FDIVF64<64, false, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDC, 7) | 0x00, 8, &OpDispatchBuilder::FDIVF64<64, false, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+      {OPD(0xDC, 0xC0), 8, &OpDispatchBuilder::FADDF64<80, false, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDC, 0xC8), 8, &OpDispatchBuilder::FMULF64<80, false, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDC, 0xE0), 8, &OpDispatchBuilder::FSUBF64<80, false, false, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDC, 0xE8), 8, &OpDispatchBuilder::FSUBF64<80, false, true, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDC, 0xF0), 8, &OpDispatchBuilder::FDIVF64<80, false, false, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDC, 0xF8), 8, &OpDispatchBuilder::FDIVF64<80, false, true, OpDispatchBuilder::OpResult::RES_STI>},
+
+    {OPDReg(0xDD, 0) | 0x00, 8, &OpDispatchBuilder::FLDF64<64>},
+
+    {OPDReg(0xDD, 1) | 0x00, 8, &OpDispatchBuilder::FISTF64<true>},
+
+    {OPDReg(0xDD, 2) | 0x00, 8, &OpDispatchBuilder::FSTF64<64>},
+
+    {OPDReg(0xDD, 3) | 0x00, 8, &OpDispatchBuilder::FSTF64<64>},
+
+    {OPDReg(0xDD, 4) | 0x00, 8, &OpDispatchBuilder::X87FRSTORF64},
+
+    // 5 = Invalid
+    {OPDReg(0xDD, 6) | 0x00, 8, &OpDispatchBuilder::X87FNSAVEF64},
+
+    {OPDReg(0xDD, 7) | 0x00, 8, &OpDispatchBuilder::X87FNSTSW},
+
+      {OPD(0xDD, 0xC0), 8, &OpDispatchBuilder::X87FFREE},
+      {OPD(0xDD, 0xD0), 8, &OpDispatchBuilder::FST}, //register-register from regular X87
+      {OPD(0xDD, 0xD8), 8, &OpDispatchBuilder::FST}, //^
+
+      {OPD(0xDD, 0xE0), 8, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+      {OPD(0xDD, 0xE8), 8, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xDE, 0) | 0x00, 8, &OpDispatchBuilder::FADDF64<16, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDE, 1) | 0x00, 8, &OpDispatchBuilder::FMULF64<16, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDE, 2) | 0x00, 8, &OpDispatchBuilder::FCOMIF64<16, true, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xDE, 3) | 0x00, 8, &OpDispatchBuilder::FCOMIF64<16, true, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, false>},
+
+    {OPDReg(0xDE, 4) | 0x00, 8, &OpDispatchBuilder::FSUBF64<16, true, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDE, 5) | 0x00, 8, &OpDispatchBuilder::FSUBF64<16, true, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDE, 6) | 0x00, 8, &OpDispatchBuilder::FDIVF64<16, true, false, OpDispatchBuilder::OpResult::RES_ST0>},
+
+    {OPDReg(0xDE, 7) | 0x00, 8, &OpDispatchBuilder::FDIVF64<16, true, true, OpDispatchBuilder::OpResult::RES_ST0>},
+
+      {OPD(0xDE, 0xC0), 8, &OpDispatchBuilder::FADDF64<80, false, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDE, 0xC8), 8, &OpDispatchBuilder::FMULF64<80, false, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDE, 0xD9), 1, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_X87, true>},
+      {OPD(0xDE, 0xE0), 8, &OpDispatchBuilder::FSUBF64<80, false, false, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDE, 0xE8), 8, &OpDispatchBuilder::FSUBF64<80, false, true, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDE, 0xF0), 8, &OpDispatchBuilder::FDIVF64<80, false, false, OpDispatchBuilder::OpResult::RES_STI>},
+      {OPD(0xDE, 0xF8), 8, &OpDispatchBuilder::FDIVF64<80, false, true, OpDispatchBuilder::OpResult::RES_STI>},
+
+    {OPDReg(0xDF, 0) | 0x00, 8, &OpDispatchBuilder::FILDF64},
+
+    {OPDReg(0xDF, 1) | 0x00, 8, &OpDispatchBuilder::FISTF64<true>},
+
+    {OPDReg(0xDF, 2) | 0x00, 8, &OpDispatchBuilder::FISTF64<false>},
+
+    {OPDReg(0xDF, 3) | 0x00, 8, &OpDispatchBuilder::FISTF64<false>},
+
+    {OPDReg(0xDF, 4) | 0x00, 8, &OpDispatchBuilder::FBLDF64},
+
+    {OPDReg(0xDF, 5) | 0x00, 8, &OpDispatchBuilder::FILDF64},
+
+    {OPDReg(0xDF, 6) | 0x00, 8, &OpDispatchBuilder::FBSTPF64},
+
+    {OPDReg(0xDF, 7) | 0x00, 8, &OpDispatchBuilder::FISTF64<false>},
+
+      // XXX: This should also set the x87 tag bits to empty
+      // We don't support this currently, so just pop the stack
+      {OPD(0xDF, 0xC0), 8, &OpDispatchBuilder::X87ModifySTP<true>},
+
+      {OPD(0xDF, 0xE0), 8, &OpDispatchBuilder::X87FNSTSW},
+      {OPD(0xDF, 0xE8), 8, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_RFLAGS, false>},
+      {OPD(0xDF, 0xF0), 8, &OpDispatchBuilder::FCOMIF64<80, false, OpDispatchBuilder::FCOMIFlags::FLAGS_RFLAGS, false>},
+  };
+  
   constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> X87OpTable[] = {
     {OPDReg(0xD8, 0) | 0x00, 8, &OpDispatchBuilder::FADD<32, false, OpDispatchBuilder::OpResult::RES_ST0>},
 
@@ -6110,7 +6463,6 @@ constexpr uint16_t PF_F2 = 3;
 #define OPD(prefix, opcode) (((prefix) << 8) | opcode)
   constexpr uint16_t PF_38_NONE = 0;
   constexpr uint16_t PF_38_66   = (1U << 0);
-  constexpr uint16_t PF_38_F2   = (1U << 1);
   constexpr uint16_t PF_38_F3   = (1U << 2);
 
   constexpr std::tuple<uint16_t, uint8_t, FEXCore::X86Tables::OpDispatchPtr> H0F38Table[] = {
@@ -6156,7 +6508,7 @@ constexpr uint16_t PF_F2 = 3;
     {OPD(PF_38_66,   0x25), 1, &OpDispatchBuilder::ExtendVectorElements<4, 8, true>},
     {OPD(PF_38_66,   0x28), 1, &OpDispatchBuilder::PMULLOp<4, true>},
     {OPD(PF_38_66,   0x29), 1, &OpDispatchBuilder::VectorALUOp<IR::OP_VCMPEQ, 8>},
-    {OPD(PF_38_66,   0x2A), 1, &OpDispatchBuilder::MOVAPSOp},
+    {OPD(PF_38_66,   0x2A), 1, &OpDispatchBuilder::MOVVectorNTOp},
     {OPD(PF_38_66,   0x2B), 1, &OpDispatchBuilder::PACKUSOp<4>},
     {OPD(PF_38_66,   0x30), 1, &OpDispatchBuilder::ExtendVectorElements<1, 2, false>},
     {OPD(PF_38_66,   0x31), 1, &OpDispatchBuilder::ExtendVectorElements<1, 4, false>},
@@ -6176,24 +6528,13 @@ constexpr uint16_t PF_F2 = 3;
     {OPD(PF_38_66,   0x40), 1, &OpDispatchBuilder::VectorALUOp<IR::OP_VSMUL, 4>},
     {OPD(PF_38_66,   0x41), 1, &OpDispatchBuilder::PHMINPOSUWOp},
 
-    {OPD(PF_38_66, 0xDB), 1, &OpDispatchBuilder::AESImcOp},
-    {OPD(PF_38_66, 0xDC), 1, &OpDispatchBuilder::AESEncOp},
-    {OPD(PF_38_66, 0xDD), 1, &OpDispatchBuilder::AESEncLastOp},
-    {OPD(PF_38_66, 0xDE), 1, &OpDispatchBuilder::AESDecOp},
-    {OPD(PF_38_66, 0xDF), 1, &OpDispatchBuilder::AESDecLastOp},
-
     {OPD(PF_38_NONE, 0xF0), 2, &OpDispatchBuilder::MOVBEOp},
     {OPD(PF_38_66, 0xF0), 2, &OpDispatchBuilder::MOVBEOp},
 
-    {OPD(PF_38_F2, 0xF0), 1, &OpDispatchBuilder::CRC32},
-    {OPD(PF_38_F2, 0xF1), 1, &OpDispatchBuilder::CRC32},
-
-    {OPD(PF_38_66 | PF_38_F2, 0xF0), 1, &OpDispatchBuilder::CRC32},
-    {OPD(PF_38_66 | PF_38_F2, 0xF1), 1, &OpDispatchBuilder::CRC32},
-
     {OPD(PF_38_66, 0xF6), 1, &OpDispatchBuilder::ADXOp},
     {OPD(PF_38_F3, 0xF6), 1, &OpDispatchBuilder::ADXOp},
   };
+
 #undef OPD
 
 #define OPD(REX, prefix, opcode) ((REX << 9) | (prefix << 8) | opcode)
@@ -6225,8 +6566,9 @@ constexpr uint16_t PF_F2 = 3;
     {OPD(0, PF_3A_66,   0x40), 1, &OpDispatchBuilder::DPPOp<4>},
     {OPD(0, PF_3A_66,   0x41), 1, &OpDispatchBuilder::DPPOp<8>},
     {OPD(0, PF_3A_66,   0x42), 1, &OpDispatchBuilder::MPSADBWOp},
+    {OPD(0, PF_3A_66,   0x44), 1, &OpDispatchBuilder::PCLMULQDQOp},
 
-    {OPD(0, PF_3A_66,   0xDF), 1, &OpDispatchBuilder::AESKeyGenAssist},
+    {OPD(0, PF_3A_NONE, 0xCC), 1, &OpDispatchBuilder::SHA1RNDS4Op},
   };
 #undef PF_3A_NONE
 #undef PF_3A_66
@@ -6308,6 +6650,8 @@ constexpr uint16_t PF_F2 = 3;
     {OPD(2, 0b10, 0xF7), 1, &OpDispatchBuilder::BMI2Shift},
     {OPD(2, 0b11, 0xF7), 1, &OpDispatchBuilder::BMI2Shift},
 
+    {OPD(3, 0b01, 0x44), 1, &OpDispatchBuilder::VPCLMULQDQOp},
+
     {OPD(3, 0b11, 0xF0), 1, &OpDispatchBuilder::RORX},
   };
 #undef OPD
@@ -6372,10 +6716,18 @@ constexpr uint16_t PF_F2 = 3;
   InstallToTable(FEXCore::X86Tables::RepNEModOps, RepNEModOpTable);
   InstallToTable(FEXCore::X86Tables::OpSizeModOps, OpSizeModOpTable);
   InstallToTable(FEXCore::X86Tables::SecondInstGroupOps, SecondaryExtensionOpTable);
+  if (Mode == Context::MODE_64BIT) {
+    InstallToTable(FEXCore::X86Tables::SecondInstGroupOps, SecondaryExtensionOpTable_64);
+  }
 
   InstallToTable(FEXCore::X86Tables::SecondModRMTableOps, SecondaryModRMExtensionOpTable);
 
-  InstallToX87Table(FEXCore::X86Tables::X87Ops, X87OpTable);
+  FEX_CONFIG_OPT(ReducedPrecision, X87REDUCEDPRECISION);
+  if(ReducedPrecision) {
+    InstallToX87Table(FEXCore::X86Tables::X87Ops, X87F64OpTable);
+  } else {
+    InstallToX87Table(FEXCore::X86Tables::X87Ops, X87OpTable);
+  }
 
   InstallToTable(FEXCore::X86Tables::H0F38TableOps, H0F38Table);
   InstallToTable(FEXCore::X86Tables::H0F3ATableOps, H0F3ATable);

External/FEXCore/Source/Interface/Core/OpcodeDispatcher.h

@@ -76,8 +76,10 @@ public:
   OrderedNode* flagsOpSrcSigned{};
 
   FEXCore::Context::Context *CTX{};
+  
+  // Used during new op bringup
   bool ShouldDump {false};
-
+  
   struct JumpTargetInfo {
     OrderedNode* BlockEntry;
     bool HaveEmitted;
@@ -148,6 +150,7 @@ public:
   }
 
   OpDispatchBuilder(FEXCore::Context::Context *ctx);
+  OpDispatchBuilder(FEXCore::Utils::IntrusivePooledAllocator &Allocator);
 
   void ResetWorkingList();
   void ResetDecodeFailure() { DecodeFailure = false; }
@@ -161,7 +164,9 @@ public:
   void UnhandledOp(OpcodeArgs);
   template<uint32_t SrcIndex>
   void MOVGPROp(OpcodeArgs);
+  void MOVGPRNTOp(OpcodeArgs);
   void MOVVectorOp(OpcodeArgs);
+  void MOVVectorNTOp(OpcodeArgs);
   void ALUOp(OpcodeArgs);
   void INTOp(OpcodeArgs);
   void SyscallOp(OpcodeArgs);
@@ -463,6 +468,57 @@ public:
   template<size_t width, bool Integer, FCOMIFlags whichflags, bool poptwice>
   void FCOMI(OpcodeArgs);
 
+  // F64 X87 Ops
+  template<size_t width>
+  void FLDF64(OpcodeArgs);
+  template<uint64_t num>
+  void FLDF64_Const(OpcodeArgs);
+
+  void FBLDF64(OpcodeArgs);
+  void FBSTPF64(OpcodeArgs);
+
+  void FILDF64(OpcodeArgs);
+
+  template<size_t width>
+  void FSTF64(OpcodeArgs);
+
+  void FSTF64(OpcodeArgs);
+
+  template<bool Truncate>
+  void FISTF64(OpcodeArgs);
+
+  template<size_t width, bool Integer, OpResult ResInST0>
+  void FADDF64(OpcodeArgs);
+  template<size_t width, bool Integer, OpResult ResInST0>
+  void FMULF64(OpcodeArgs);
+  template<size_t width, bool Integer, bool reverse, OpResult ResInST0>
+  void FDIVF64(OpcodeArgs);
+  template<size_t width, bool Integer, bool reverse, OpResult ResInST0>
+  void FSUBF64(OpcodeArgs);
+  void FCHSF64(OpcodeArgs);
+  void FABSF64(OpcodeArgs);
+  void FTSTF64(OpcodeArgs);
+  void FRNDINTF64(OpcodeArgs);
+  void FXTRACTF64(OpcodeArgs);
+  void FNINITF64(OpcodeArgs);
+  void FSQRTF64(OpcodeArgs);
+  template<FEXCore::IR::IROps IROp>
+  void X87UnaryOpF64(OpcodeArgs);
+  template<FEXCore::IR::IROps IROp>
+  void X87BinaryOpF64(OpcodeArgs);
+  void X87SinCosF64(OpcodeArgs);
+  void X87FLDCWF64(OpcodeArgs);
+  void X87FYL2XF64(OpcodeArgs);
+  void X87TANF64(OpcodeArgs);
+  void X87ATANF64(OpcodeArgs);
+  void X87FNSAVEF64(OpcodeArgs);
+  void X87FRSTORF64(OpcodeArgs);
+  void X87FXAMF64(OpcodeArgs);
+  void X87LDENVF64(OpcodeArgs);
+  
+  template<size_t width, bool Integer, FCOMIFlags whichflags, bool poptwice>
+  void FCOMIF64(OpcodeArgs);
+
   void FXSaveOp(OpcodeArgs);
   void FXRStoreOp(OpcodeArgs);
 
@@ -535,6 +591,15 @@ public:
 
   void PSADBW(OpcodeArgs);
 
+  void SHA1NEXTEOp(OpcodeArgs);
+  void SHA1MSG1Op(OpcodeArgs);
+  void SHA1MSG2Op(OpcodeArgs);
+  void SHA1RNDS4Op(OpcodeArgs);
+
+  void SHA256MSG1Op(OpcodeArgs);
+  void SHA256MSG2Op(OpcodeArgs);
+  void SHA256RNDS2Op(OpcodeArgs);
+
   void AESImcOp(OpcodeArgs);
   void AESEncOp(OpcodeArgs);
   void AESEncLastOp(OpcodeArgs);
@@ -558,6 +623,8 @@ public:
   void DPPOp(OpcodeArgs);
 
   void MPSADBWOp(OpcodeArgs);
+  void PCLMULQDQOp(OpcodeArgs);
+  void VPCLMULQDQOp(OpcodeArgs);
 
   void CRC32(OpcodeArgs);
 
@@ -580,12 +647,22 @@ private:
 
   OrderedNode *AppendSegmentOffset(OrderedNode *Value, uint32_t Flags, uint32_t DefaultPrefix = 0, bool Override = false);
 
+  enum class MemoryAccessType {
+    // Choose TSO or Non-TSO depending on access type
+    ACCESS_DEFAULT,
+    // TSO access behaviour
+    ACCESS_TSO,
+    // Non-TSO access behaviour
+    ACCESS_NONTSO,
+    // Non-temporal streaming
+    ACCESS_STREAM,
+  };
   OrderedNode *GetRelocatedPC(FEXCore::X86Tables::DecodedOp const& Op, int64_t Offset = 0);
-  OrderedNode *LoadSource(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint32_t Flags, int8_t Align, bool LoadData = true, bool ForceLoad = false);
-  OrderedNode *LoadSource_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint8_t OpSize, uint32_t Flags, int8_t Align, bool LoadData = true, bool ForceLoad = false);
-  void StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, FEXCore::X86Tables::DecodedOperand const& Operand, OrderedNode *const Src, uint8_t OpSize, int8_t Align);
-  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, FEXCore::X86Tables::DecodedOperand const& Operand, OrderedNode *const Src, int8_t Align);
-  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, OrderedNode *const Src, int8_t Align);
+  OrderedNode *LoadSource(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint32_t Flags, int8_t Align, bool LoadData = true, bool ForceLoad = false, MemoryAccessType AccessType = MemoryAccessType::ACCESS_DEFAULT);
+  OrderedNode *LoadSource_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint8_t OpSize, uint32_t Flags, int8_t Align, bool LoadData = true, bool ForceLoad = false, MemoryAccessType AccessType = MemoryAccessType::ACCESS_DEFAULT);
+  void StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, FEXCore::X86Tables::DecodedOperand const& Operand, OrderedNode *const Src, uint8_t OpSize, int8_t Align, MemoryAccessType AccessType = MemoryAccessType::ACCESS_DEFAULT);
+  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, FEXCore::X86Tables::DecodedOperand const& Operand, OrderedNode *const Src, int8_t Align, MemoryAccessType AccessType = MemoryAccessType::ACCESS_DEFAULT);
+  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, OrderedNode *const Src, int8_t Align, MemoryAccessType AccessType = MemoryAccessType::ACCESS_DEFAULT);
 
   [[nodiscard]] static uint32_t GPROffset(X86State::X86Reg reg) {
     LOGMAN_THROW_A_FMT(reg <= X86State::X86Reg::REG_R15, "Invalid reg used");
@@ -1163,18 +1240,20 @@ private:
   uint64_t Entry;
 
   OrderedNode* _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, uint8_t Size, OrderedNode *Addr, OrderedNode *Value, uint8_t Align = 1) {
-    if (CTX->Config.TSOEnabled)
+    if (CTX->IsTSOEnabled())
       return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
     else
       return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
   }
 
   OrderedNode* _LoadMemAutoTSO(FEXCore::IR::RegisterClassType Class, uint8_t Size, OrderedNode *ssa0, uint8_t Align = 1) {
-    if (CTX->Config.TSOEnabled)
+    if (CTX->IsTSOEnabled())
       return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
     else
       return _LoadMem(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
   }
+
+  void InstallHostSpecificOpcodeHandlers();
 };
 
 void InstallOpcodeHandlers(Context::OperatingMode Mode);

External/FEXCore/Source/Interface/Core/OpcodeDispatcher/Crypto.cpp

@@ -10,13 +10,259 @@ $end_info$
 #include <FEXCore/Utils/LogManager.h>
 #include "Interface/Core/OpcodeDispatcher.h"
 
-#include <stdint.h>
+#include <array>
+#include <cstdint>
+#include <tuple>
+#include <utility>
 
 namespace FEXCore::IR {
 class OrderedNode;
 
 #define OpcodeArgs [[maybe_unused]] FEXCore::X86Tables::DecodedOp Op
 
+void OpDispatchBuilder::SHA1NEXTEOp(OpcodeArgs) {
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  auto Tmp = _Ror(_VExtractToGPR(16, 4, Dest, 3), _Constant(32, 2));
+  auto Top = _Add(_VExtractToGPR(16, 4, Src, 3), Tmp);
+  auto Result = _VInsGPR(16, 4, 3, Src, Top);
+
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::SHA1MSG1Op(OpcodeArgs) {
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  auto W0 = _VExtractToGPR(16, 4, Dest, 3);
+  auto W1 = _VExtractToGPR(16, 4, Dest, 2);
+  auto W2 = _VExtractToGPR(16, 4, Dest, 1);
+  auto W3 = _VExtractToGPR(16, 4, Dest, 0);
+  auto W4 = _VExtractToGPR(16, 4, Src, 3);
+  auto W5 = _VExtractToGPR(16, 4, Src, 2);
+
+  auto D3 = _VInsGPR(16, 4, 3, Dest, _Xor(W2, W0));
+  auto D2 = _VInsGPR(16, 4, 2, D3,   _Xor(W3, W1));
+  auto D1 = _VInsGPR(16, 4, 1, D2,   _Xor(W4, W2));
+  auto D0 = _VInsGPR(16, 4, 0, D1,   _Xor(W5, W3));
+
+  StoreResult(FPRClass, Op, D0, -1);
+}
+
+void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  // ROR by 31 is equivalent to a ROL by 1
+  auto ThirtyOne = _Constant(32, 31);
+
+  auto W13 = _VExtractToGPR(16, 4, Src, 2);
+  auto W14 = _VExtractToGPR(16, 4, Src, 1);
+  auto W15 = _VExtractToGPR(16, 4, Src, 0);
+  auto W16 = _Ror(_Xor(_VExtractToGPR(16, 4, Dest, 3), W13), ThirtyOne);
+  auto W17 = _Ror(_Xor(_VExtractToGPR(16, 4, Dest, 2), W14), ThirtyOne);
+  auto W18 = _Ror(_Xor(_VExtractToGPR(16, 4, Dest, 1), W15), ThirtyOne);
+  auto W19 = _Ror(_Xor(_VExtractToGPR(16, 4, Dest, 0), W16), ThirtyOne);
+
+  auto D3 = _VInsGPR(16, 4, 3, Dest, W16);
+  auto D2 = _VInsGPR(16, 4, 2, D3, W17);
+  auto D1 = _VInsGPR(16, 4, 1, D2, W18);
+  auto D0 = _VInsGPR(16, 4, 0, D1, W19);
+
+  StoreResult(FPRClass, Op, D0, -1);
+}
+
+void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
+  LOGMAN_THROW_A_FMT(Op->Src[1].IsLiteral(),
+                     "Src1 needs to be literal here to indicate function and constants");
+
+  using FnType = OrderedNode* (*)(OpDispatchBuilder&, OrderedNode*, OrderedNode*, OrderedNode*);
+
+  const auto f0 = [](OpDispatchBuilder &Self, OrderedNode *B, OrderedNode *C, OrderedNode *D) -> OrderedNode* {
+    return Self._Xor(Self._And(B, C), Self._And(Self._Not(B), D));
+  };
+  const auto f1 = [](OpDispatchBuilder &Self, OrderedNode *B, OrderedNode *C, OrderedNode *D) -> OrderedNode* {
+    return Self._Xor(Self._Xor(B, C), D);
+  };
+  const auto f2 = [](OpDispatchBuilder &Self, OrderedNode *B, OrderedNode *C, OrderedNode *D) -> OrderedNode* {
+    return Self._Xor(Self._Xor(Self._And(B, C), Self._And(B, D)), Self._And(C, D));
+  };
+  const auto f3 = [](OpDispatchBuilder &Self, OrderedNode *B, OrderedNode *C, OrderedNode *D) -> OrderedNode* {
+    return Self._Xor(Self._Xor(B, C), D);
+  };
+
+  constexpr std::array<uint32_t, 4> k_array{
+    0x5A827999U,
+    0x6ED9EBA1U,
+    0x8F1BBCDCU,
+    0xCA62C1D6U,
+  };
+
+  constexpr std::array<FnType, 4> fn_array{
+    f0, f1, f2, f3,
+  };
+
+  const uint64_t Imm8 = Op->Src[1].Data.Literal.Value & 0b11;
+  const FnType Fn = fn_array[Imm8];
+  auto K = _Constant(32, k_array[Imm8]);
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  auto W0E = _VExtractToGPR(16, 4, Src, 3);
+  auto W1  = _VExtractToGPR(16, 4, Src, 2);
+  auto W2  = _VExtractToGPR(16, 4, Src, 1);
+  auto W3  = _VExtractToGPR(16, 4, Src, 0);
+
+  using RoundResult = std::tuple<OrderedNode*, OrderedNode*, OrderedNode*, OrderedNode*, OrderedNode*>;
+
+  const auto Round0 = [&]() -> RoundResult {
+    auto A = _VExtractToGPR(16, 4, Dest, 3);
+    auto B = _VExtractToGPR(16, 4, Dest, 2);
+    auto C = _VExtractToGPR(16, 4, Dest, 1);
+    auto D = _VExtractToGPR(16, 4, Dest, 0);
+
+    auto A1 = _Add(_Add(_Add(Fn(*this, B, C, D), _Ror(A, _Constant(32, 27))), W0E), K);
+    auto B1 = A;
+    auto C1 = _Ror(B, _Constant(32, 2));
+    auto D1 = C;
+    auto E1 = D;
+
+    return {A1, B1, C1, D1, E1};
+  };
+  const auto Round1To3 = [&](OrderedNode *A, OrderedNode *B, OrderedNode *C,
+                             OrderedNode *D, OrderedNode *E, OrderedNode *W) -> RoundResult {
+    auto ANext = _Add(_Add(_Add(_Add(Fn(*this, B, C, D), _Ror(A, _Constant(32, 27))), W), E), K);
+    auto BNext = A;
+    auto CNext = _Ror(B, _Constant(32, 2));
+    auto DNext = C;
+    auto ENext = D;
+
+    return {ANext, BNext, CNext, DNext, ENext};
+  };
+
+  auto [A1, B1, C1, D1, E1] = Round0();
+  auto [A2, B2, C2, D2, E2] = Round1To3(A1, B1, C1, D1, E1, W1);
+  auto [A3, B3, C3, D3, E3] = Round1To3(A2, B2, C2, D2, E2, W2);
+  auto Final                = Round1To3(A3, B3, C3, D3, E3, W3);
+
+  auto Dest3 = _VInsGPR(16, 4, 3, Dest,  std::get<0>(Final));
+  auto Dest2 = _VInsGPR(16, 4, 2, Dest3, std::get<1>(Final));
+  auto Dest1 = _VInsGPR(16, 4, 1, Dest2, std::get<2>(Final));
+  auto Dest0 = _VInsGPR(16, 4, 0, Dest1, std::get<3>(Final));
+
+  StoreResult(FPRClass, Op, Dest0, -1);
+}
+
+void OpDispatchBuilder::SHA256MSG1Op(OpcodeArgs) {
+  const auto Sigma0 = [this](OrderedNode* W) -> OrderedNode* {
+    return _Xor(_Xor(_Ror(W, _Constant(32, 7)), _Ror(W, _Constant(32, 18))), _Lshr(W, _Constant(32, 3)));
+  };
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  auto W4 = _VExtractToGPR(16, 4, Src, 0);
+  auto W3 = _VExtractToGPR(16, 4, Dest, 3);
+  auto W2 = _VExtractToGPR(16, 4, Dest, 2);
+  auto W1 = _VExtractToGPR(16, 4, Dest, 1);
+  auto W0 = _VExtractToGPR(16, 4, Dest, 0);
+
+  auto Sig3 = _Add(W3, Sigma0(W4));
+  auto Sig2 = _Add(W2, Sigma0(W3));
+  auto Sig1 = _Add(W1, Sigma0(W2));
+  auto Sig0 = _Add(W0, Sigma0(W1));
+
+  auto D3 = _VInsGPR(16, 4, 3, Dest, Sig3);
+  auto D2 = _VInsGPR(16, 4, 2, D3, Sig2);
+  auto D1 = _VInsGPR(16, 4, 1, D2, Sig1);
+  auto D0 = _VInsGPR(16, 4, 0, D1, Sig0);
+
+  StoreResult(FPRClass, Op, D0, -1);
+}
+
+void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {
+  const auto Sigma1 = [this](OrderedNode* W) -> OrderedNode* {
+    return _Xor(_Xor(_Ror(W, _Constant(32, 17)), _Ror(W, _Constant(32, 19))), _Lshr(W, _Constant(32, 10)));
+  };
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  auto W14 = _VExtractToGPR(16, 4, Src, 2);
+  auto W15 = _VExtractToGPR(16, 4, Src, 3);
+  auto W16 = _Add(_VExtractToGPR(16, 4, Dest, 0), Sigma1(W14));
+  auto W17 = _Add(_VExtractToGPR(16, 4, Dest, 1), Sigma1(W15));
+  auto W18 = _Add(_VExtractToGPR(16, 4, Dest, 2), Sigma1(W16));
+  auto W19 = _Add(_VExtractToGPR(16, 4, Dest, 3), Sigma1(W17));
+
+  auto D3 = _VInsGPR(16, 4, 3, Dest, W19);
+  auto D2 = _VInsGPR(16, 4, 2, D3, W18);
+  auto D1 = _VInsGPR(16, 4, 1, D2, W17);
+  auto D0 = _VInsGPR(16, 4, 0, D1, W16);
+
+  StoreResult(FPRClass, Op, D0, -1);
+}
+
+void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
+  const auto Ch = [this](OrderedNode *E, OrderedNode *F, OrderedNode *G) -> OrderedNode* {
+    return _Xor(_And(E, F), _And(_Not(E), G));
+  };
+  const auto Major = [this](OrderedNode *A, OrderedNode *B, OrderedNode *C) -> OrderedNode* {
+    return _Xor(_Xor(_And(A, B), _And(A, C)), _And(B, C));
+  };
+  const auto Sigma0 = [this](OrderedNode *A) -> OrderedNode* {
+    return _Xor(_Xor(_Ror(A, _Constant(32, 2)), _Ror(A, _Constant(32, 13))), _Ror(A, _Constant(32, 22)));
+  };
+  const auto Sigma1 = [this](OrderedNode *E) -> OrderedNode* {
+    return _Xor(_Xor(_Ror(E, _Constant(32, 6)), _Ror(E, _Constant(32, 11))), _Ror(E, _Constant(32, 25)));
+  };
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  OrderedNode *XMM0 = _LoadContext(16, FPRClass, offsetof(FEXCore::Core::CPUState, xmm[0]));
+
+  auto A0 = _VExtractToGPR(16, 4, Src, 3);
+  auto B0 = _VExtractToGPR(16, 4, Src, 2);
+  auto C0 = _VExtractToGPR(16, 4, Dest, 3);
+  auto D0 = _VExtractToGPR(16, 4, Dest, 2);
+  auto E0 = _VExtractToGPR(16, 4, Src, 1);
+  auto F0 = _VExtractToGPR(16, 4, Src, 0);
+  auto G0 = _VExtractToGPR(16, 4, Dest, 1);
+  auto H0 = _VExtractToGPR(16, 4, Dest, 0);
+  auto WK0 = _VExtractToGPR(16, 4, XMM0, 0);
+  auto WK1 = _VExtractToGPR(16, 4, XMM0, 1);
+
+  using RoundResult = std::tuple<OrderedNode*, OrderedNode*, OrderedNode*, OrderedNode*,
+                                 OrderedNode*, OrderedNode*, OrderedNode*, OrderedNode*>;
+  const auto Round = [&](OrderedNode *A, OrderedNode *B, OrderedNode *C, OrderedNode *D,
+                         OrderedNode *E, OrderedNode *F, OrderedNode *G, OrderedNode *H,
+                         OrderedNode* WK) -> RoundResult {
+    auto ANext = _Add(_Add(_Add(_Add(_Add(Ch(E, F, G), Sigma1(E)), WK), H), Major(A, B, C)), Sigma0(A));
+    auto BNext = A;
+    auto CNext = B;
+    auto DNext = C;
+    auto ENext = _Add(_Add(_Add(_Add(Ch(E, F, G), Sigma1(E)), WK), H), D);
+    auto FNext = E;
+    auto GNext = F;
+    auto HNext = G;
+
+    return {ANext, BNext, CNext, DNext, ENext, FNext, GNext, HNext};
+  };
+
+
+  auto [A1, B1, C1, D1, E1, F1, G1, H1] = Round(A0, B0, C0, D0, E0, F0, G0, H0, WK0);
+  auto Final                            = Round(A1, B1, C1, D1, E1, F1, G1, H1, WK1);
+
+  auto Res3 = _VInsGPR(16, 4, 3, Dest, std::get<0>(Final));
+  auto Res2 = _VInsGPR(16, 4, 2, Res3, std::get<1>(Final));
+  auto Res1 = _VInsGPR(16, 4, 1, Res2, std::get<4>(Final));
+  auto Res0 = _VInsGPR(16, 4, 0, Res1, std::get<5>(Final));
+
+  StoreResult(FPRClass, Op, Res0, -1);
+}
+
 void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
   auto Res = _VAESImc(Src);
@@ -60,4 +306,26 @@ void OpDispatchBuilder::AESKeyGenAssist(OpcodeArgs) {
   StoreResult(FPRClass, Op, Res, -1);
 }
 
+void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
+  LOGMAN_THROW_A_FMT(Op->Src[1].IsLiteral(), "Selector needs to be literal here");
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  const auto Selector = static_cast<uint8_t>(Op->Src[1].Data.Literal.Value);
+
+  auto Res = _PCLMUL(Dest, Src, Selector);
+  StoreResult(FPRClass, Op, Res, -1);
+}
+
+void OpDispatchBuilder::VPCLMULQDQOp(OpcodeArgs) {
+  LOGMAN_THROW_A_FMT(Op->Src[2].IsLiteral(), "Selector needs to be literal here");
+
+  OrderedNode *Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  OrderedNode *Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags, -1);
+  const auto Selector = static_cast<uint8_t>(Op->Src[2].Data.Literal.Value);
+
+  auto Res = _PCLMUL(Src1, Src2, Selector);
+  StoreResult(FPRClass, Op, Res, -1);
+}
+
 }

External/FEXCore/Source/Interface/Core/OpcodeDispatcher/Vector.cpp

@@ -28,6 +28,11 @@ void OpDispatchBuilder::MOVVectorOp(OpcodeArgs) {
   StoreResult(FPRClass, Op, Src, 1);
 }
 
+void OpDispatchBuilder::MOVVectorNTOp(OpcodeArgs) {
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, 1, true, false, MemoryAccessType::ACCESS_STREAM);
+  StoreResult(FPRClass, Op, Src, 1, MemoryAccessType::ACCESS_STREAM);
+}
+
 void OpDispatchBuilder::MOVAPSOp(OpcodeArgs) {
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
   StoreResult(FPRClass, Op, Src, -1);
@@ -750,7 +755,7 @@ void OpDispatchBuilder::PExtrOp(OpcodeArgs) {
   }
   else {
     // If we are storing to memory then we store the size of the element extracted
-    StoreResult(GPRClass, Op, Result, -1);
+    StoreResult_WithOpSize(GPRClass, Op, Op->Dest, Result, ElementSize, -1);
   }
 }
 
@@ -1194,7 +1199,8 @@ void OpDispatchBuilder::MASKMOVOp(OpcodeArgs) {
       {
         auto DestByte = _Bfe(8, 8 * Select, DestElement);
         auto MemLocation = _Add(MemDest, _Constant(Element * 8 + Select));
-        _StoreMemAutoTSO(GPRClass, 1, MemLocation, DestByte, 1);
+        // MASKMOVDQU/MASKMOVQ is explicitly weakly-ordered on its store
+        _StoreMem(GPRClass, 1, MemLocation, DestByte, 1);
       }
       auto Jump = _Jump();
       auto NextJumpTarget = CreateNewCodeBlockAfter(StoreBlock);
@@ -1809,7 +1815,6 @@ void OpDispatchBuilder::VPFCMPOp(OpcodeArgs) {
   }
 
   StoreResult(FPRClass, Op, Result, -1);
-  ShouldDump = true;
 }
 
 template

External/FEXCore/Source/Interface/Core/OpcodeDispatcher/X87.cpp

@@ -621,8 +621,8 @@ void OpDispatchBuilder::FXTRACT(OpcodeArgs) {
 }
 
 void OpDispatchBuilder::FNINIT(OpcodeArgs) {
-  // Init FCW to 0x037
-  auto NewFCW = _Constant(16, 0x037);
+  // Init FCW to 0x037F
+  auto NewFCW = _Constant(16, 0x037F);
   _F80LoadFCW(NewFCW);
   _StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 

External/FEXCore/Source/Interface/Core/SignalDelegator.cpp

@@ -95,10 +95,11 @@ namespace FEXCore {
       LogMan::Msg::EFmt("[{}] Thread has received a signal and hasn't registered itself with the delegate! Programming error!", FHU::Syscalls::gettid());
     }
     else {
-      if (Handler.Handler &&
-          Handler.Handler(Thread, Signal, Info, UContext)) {
-        // If the host handler handled the fault then we can continue now
-        return;
+      for (auto &Handler : Handler.Handlers) {
+        if (Handler(Thread, Signal, Info, UContext)) {
+          // If the host handler handled the fault then we can continue now
+          return;
+        }
       }
 
       if (Handler.FrontendHandler &&

External/FEXCore/Source/Interface/Core/X86Tables/BaseTables.cpp

@@ -167,7 +167,7 @@ void InitializeBaseTables(Context::OperatingMode Mode) {
     {0xC2, 1, X86InstInfo{"RET",    TYPE_INST, FLAGS_SETS_RIP | FLAGS_BLOCK_END,                                             2, nullptr}},
     {0xC3, 1, X86InstInfo{"RET",    TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_SETS_RIP | FLAGS_BLOCK_END ,                                                0, nullptr}},
     {0xC8, 1, X86InstInfo{"ENTER",  TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_DEBUG_MEM_ACCESS ,                                      3, nullptr}},
-    {0xC9, 1, X86InstInfo{"LEAVE",  TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_SETS_RIP | FLAGS_BLOCK_END ,                                                0, nullptr}},
+    {0xC9, 1, X86InstInfo{"LEAVE",  TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_DEBUG_MEM_ACCESS ,                                                0, nullptr}},
     {0xCA, 2, X86InstInfo{"RETF",   TYPE_PRIV, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_SETS_RIP | FLAGS_BLOCK_END,                                                              0, nullptr}},
     {0xCC, 1, X86InstInfo{"INT3",   TYPE_INST, FLAGS_DEBUG,                                                                                      0, nullptr}},
     {0xCD, 1, X86InstInfo{"INT",    TYPE_INST, FLAGS_DEBUG ,                                                                  1, nullptr}},

External/FEXCore/Source/Interface/Core/X86Tables/H0F38Tables.cpp

@@ -87,6 +87,14 @@ void InitializeH0F38Tables() {
     {OPD(PF_38_66,   0x40), 1, X86InstInfo{"PMULLD",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(PF_38_66,   0x41), 1, X86InstInfo{"PHMINPOSUW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
+    {OPD(PF_38_NONE, 0xC8), 1, X86InstInfo{"SHA1NEXTE",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(PF_38_NONE, 0xC9), 1, X86InstInfo{"SHA1MSG1",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(PF_38_NONE, 0xCA), 1, X86InstInfo{"SHA1MSG2",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+
+    {OPD(PF_38_NONE, 0xCB), 1, X86InstInfo{"SHA256RNDS2", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(PF_38_NONE, 0xCC), 1, X86InstInfo{"SHA256MSG1",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(PF_38_NONE, 0xCD), 1, X86InstInfo{"SHA256MSG2",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+
     {OPD(PF_38_66,   0xDB), 1, X86InstInfo{"AESIMC",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(PF_38_66,   0xDC), 1, X86InstInfo{"AESENC",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(PF_38_66,   0xDD), 1, X86InstInfo{"AESENCLAST", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},

External/FEXCore/Source/Interface/Core/X86Tables/H0F3ATables.cpp

@@ -31,7 +31,7 @@ void InitializeH0F3ATables(Context::OperatingMode Mode) {
     {OPD(0, PF_3A_66,   0x0E), 1, X86InstInfo{"PBLENDW",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,           1, nullptr}},
     {OPD(0, PF_3A_66,   0x0F), 1, X86InstInfo{"PALIGNR",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
 
-    {OPD(0, PF_3A_66,   0x14), 1, X86InstInfo{"PEXTRB",          TYPE_INST, GenFlagsSizes(SIZE_8BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(0, PF_3A_66,   0x14), 1, X86InstInfo{"PEXTRB",          TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(0, PF_3A_66,   0x15), 1, X86InstInfo{"PEXTRW",          TYPE_INST, GenFlagsSizes(SIZE_16BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(0, PF_3A_66,   0x16), 1, X86InstInfo{"PEXTRD",          TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(0, PF_3A_66,   0x17), 1, X86InstInfo{"EXTRACTPS",       TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
@@ -42,13 +42,15 @@ void InitializeH0F3ATables(Context::OperatingMode Mode) {
     {OPD(0, PF_3A_66,   0x40), 1, X86InstInfo{"DPPS",            TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(0, PF_3A_66,   0x41), 1, X86InstInfo{"DPPD",            TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(0, PF_3A_66,   0x42), 1, X86InstInfo{"MPSADBW",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
-    {OPD(0, PF_3A_66,   0x44), 1, X86InstInfo{"PCLMULQDQ",       TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(0, PF_3A_66,   0x44), 1, X86InstInfo{"PCLMULQDQ",       TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
 
     {OPD(0, PF_3A_66,   0x60), 1, X86InstInfo{"PCMPESTRM",       TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(0, PF_3A_66,   0x61), 1, X86InstInfo{"PCMPESTRI",       TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(0, PF_3A_66,   0x62), 1, X86InstInfo{"PCMPISTRM",       TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(0, PF_3A_66,   0x63), 1, X86InstInfo{"PCMPISTRI",       TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
 
+    {OPD(0, PF_3A_NONE, 0xCC), 1, X86InstInfo{"SHA1RNDS4",       TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+
     {OPD(0, PF_3A_66,   0xDF), 1, X86InstInfo{"AESKEYGENASSIST", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
   };
 

External/FEXCore/Source/Interface/Core/X86Tables/SecondaryGroupTables.cpp

@@ -343,8 +343,8 @@ void InitializeSecondaryGroupTables() {
 
     {OPD(TYPE_GROUP_15, PF_F3, 0), 1, X86InstInfo{"RDFSBASE", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_REG_ONLY,          0, nullptr}},
     {OPD(TYPE_GROUP_15, PF_F3, 1), 1, X86InstInfo{"RDGSBASE", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_REG_ONLY,          0, nullptr}},
-    {OPD(TYPE_GROUP_15, PF_F3, 2), 1, X86InstInfo{"WRFSBASE", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY,          0, nullptr}},
-    {OPD(TYPE_GROUP_15, PF_F3, 3), 1, X86InstInfo{"WRGSBASE", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY,          0, nullptr}},
+    {OPD(TYPE_GROUP_15, PF_F3, 2), 1, X86InstInfo{"WRFSBASE", TYPE_INST, GenFlagsDstSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY,          0, nullptr}},
+    {OPD(TYPE_GROUP_15, PF_F3, 3), 1, X86InstInfo{"WRGSBASE", TYPE_INST, GenFlagsDstSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY,          0, nullptr}},
     {OPD(TYPE_GROUP_15, PF_F3, 4), 1, X86InstInfo{"", TYPE_INVALID, FLAGS_NONE,                    0, nullptr}},
     {OPD(TYPE_GROUP_15, PF_F3, 5), 1, X86InstInfo{"INCSSPQ", TYPE_INST, FLAGS_MODRM,                    0, nullptr}},
     {OPD(TYPE_GROUP_15, PF_F3, 6), 1, X86InstInfo{"CLRSSBSY", TYPE_INST, FLAGS_NONE,                    0, nullptr}},

External/FEXCore/Source/Interface/Core/X86Tables/VEXTables.cpp

@@ -441,7 +441,7 @@ void InitializeVEXTables() {
     {OPD(3, 0b01, 0x40), 1, X86InstInfo{"VDPPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x41), 1, X86InstInfo{"VDPPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x42), 1, X86InstInfo{"VMPSADBW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x44), 1, X86InstInfo{"VPCLMULQDQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x44), 1, X86InstInfo{"VPCLMULQDQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(3, 0b01, 0x46), 1, X86InstInfo{"VPERM2I128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
     {OPD(3, 0b01, 0x48), 1, X86InstInfo{"VPERMILzz2PS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},

External/FEXCore/Source/Interface/GDBJIT/GDBJIT.cpp

@@ -0,0 +1,127 @@
+#include "GDBJIT.h"
+
+#include <FEXCore/Debug/InternalThreadState.h>
+#include <FEXCore/HLE/SourcecodeResolver.h>
+#include <FEXCore/Utils/LogManager.h>
+
+#if defined(GDB_SYMBOLS_ENABLED)
+
+#include <FEXCore/Debug/GDBReaderInterface.h>
+
+extern "C" {
+enum jit_actions_t { JIT_NOACTION = 0, JIT_REGISTER_FN, JIT_UNREGISTER_FN };
+
+struct jit_code_entry {
+  jit_code_entry *next_entry;
+  jit_code_entry *prev_entry;
+  const char *symfile_addr;
+  uint64_t symfile_size;
+};
+
+struct jit_descriptor {
+  uint32_t version;
+  /* This type should be jit_actions_t, but we use uint32_t
+     to be explicit about the bitwidth.  */
+  uint32_t action_flag;
+  jit_code_entry *relevant_entry;
+  jit_code_entry *first_entry;
+};
+
+/* Make sure to specify the version statically, because the
+   debugger may check the version before we can set it.  */
+
+constinit jit_descriptor __jit_debug_descriptor = {.version = 1};
+
+/* GDB puts a breakpoint in this function.  */
+void __attribute__((noinline)) __jit_debug_register_code() {
+  asm volatile("" ::"r"(&__jit_debug_descriptor));
+};
+}
+
+namespace FEXCore {
+
+void GDBJITRegister(FEXCore::IR::AOTIRCacheEntry *Entry, uintptr_t VAFileStart,
+                    uint64_t GuestRIP, uintptr_t HostEntry,
+                    FEXCore::Core::DebugData *DebugData) {
+  auto map = Entry->SourcecodeMap.get();
+
+  if (map) {
+    auto FileOffset = GuestRIP - VAFileStart;
+
+    auto Sym = map->FindSymbolMapping(FileOffset);
+
+    std::string SymName = HLE::SourcecodeSymbolMapping::SymName(
+        Sym, Entry->Filename, HostEntry, FileOffset);
+
+    std::vector<gdb_line_mapping> Lines;
+    for (const auto &GuestOpcode : DebugData->GuestOpcodes) {
+      auto Line = map->FindLineMapping(GuestRIP + GuestOpcode.GuestEntryOffset -
+                                       VAFileStart);
+      if (Line) {
+        Lines.push_back(
+            {Line->LineNumber, HostEntry + GuestOpcode.HostEntryOffset});
+      }
+    }
+
+    size_t size = sizeof(info_t) + 1 * sizeof(blocks_t) +
+                  Lines.size() * sizeof(gdb_line_mapping);
+
+    auto mem = (uint8_t *)malloc(size);
+    auto base = mem;
+    info_t *info = (info_t *)mem;
+    mem += sizeof(info_t);
+
+    strncpy(info->filename, map->SourceFile.c_str(), 511);
+
+    info->nblocks = 1;
+
+    auto blocks = (blocks_t *)mem;
+    info->blocks_ofs = mem - base;
+
+    mem += info->nblocks * sizeof(blocks_t);
+
+    for (int i = 0; i < info->nblocks; i++) {
+      strncpy(blocks[i].name, SymName.c_str(), 511);
+      blocks[i].start = HostEntry;
+      blocks[i].end = HostEntry + DebugData->HostCodeSize;
+    }
+
+    info->nlines = Lines.size();
+
+    auto lines = (gdb_line_mapping *)mem;
+    info->lines_ofs = mem - base;
+    mem += info->nlines * sizeof(gdb_line_mapping);
+
+    if (info->nlines) {
+      memcpy(lines, &Lines.at(0), info->nlines * sizeof(gdb_line_mapping));
+    }
+
+    auto entry = new jit_code_entry{0, 0, 0, 0};
+
+    entry->symfile_addr = (const char *)info;
+    entry->symfile_size = size;
+
+    if (__jit_debug_descriptor.first_entry) {
+      __jit_debug_descriptor.relevant_entry->next_entry = entry;
+      entry->prev_entry = __jit_debug_descriptor.relevant_entry;
+    } else {
+      __jit_debug_descriptor.first_entry = entry;
+    }
+
+    __jit_debug_descriptor.relevant_entry = entry;
+    __jit_debug_descriptor.action_flag = JIT_REGISTER_FN;
+    __jit_debug_register_code();
+  }
+}
+} // namespace FEXCore
+#else
+namespace FEXCore {
+void GDBJITRegister([[maybe_unused]] FEXCore::IR::AOTIRCacheEntry *Entry,
+                    [[maybe_unused]] uintptr_t VAFileStart,
+                    [[maybe_unused]] uint64_t GuestRIP,
+                    [[maybe_unused]] uintptr_t HostEntry,
+                    [[maybe_unused]] FEXCore::Core::DebugData *DebugData) {
+  ERROR_AND_DIE_FMT("GDBSymbols support not compiled in");
+}
+} // namespace FEXCore
+#endif

External/FEXCore/Source/Interface/GDBJIT/GDBJIT.h

@@ -0,0 +1,7 @@
+
+
+#include <Interface/IR/AOTIR.h>
+
+namespace FEXCore {
+    void GDBJITRegister(FEXCore::IR::AOTIRCacheEntry *Entry, uintptr_t VAFileStart, uint64_t GuestRIP, uintptr_t HostEntry, FEXCore::Core::DebugData *DebugData);
+}

External/FEXCore/Source/Interface/HLE/Thunks/Thunks.cpp

@@ -10,14 +10,18 @@ $end_info$
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/IR/IR.h>
+#include <FEXCore/IR/IREmitter.h>
+#include "FEXCore/Utils/CompilerDefs.h"
 #include "Thunks.h"
 
+#include <cstdint>
 #include <dlfcn.h>
 
 #include <Interface/Context/Context.h>
 #include "FEXCore/Core/X86Enums.h"
 #include <malloc.h>
-#include <map>
+#include <mutex>
+#include <unordered_map>
 #include <memory>
 #include <shared_mutex>
 #include <stdint.h>
@@ -26,26 +30,106 @@ $end_info$
 
 struct LoadlibArgs {
     const char *Name;
-    uintptr_t CallbackThunks;
 };
 
 static thread_local FEXCore::Core::InternalThreadState *Thread;
 
+static __attribute__((aligned(16), naked, section("HostToGuestTrampolineTemplate"))) void HostToGuestTrampolineTemplate() {
+#if defined(_M_X86_64)
+  asm(
+    "lea 0f(%rip), %r11 \n"
+    "jmpq *0f(%rip) \n"
+    ".align 8 \n"
+    "0: \n"
+    ".quad 0, 0, 0, 0 \n" // TrampolineInstanceInfo
+  );
+#elif defined(_M_ARM_64)
+  asm(
+    "adr x11, 0f \n"
+    "ldr x16, [x11] \n"
+    "br x16 \n"
+    // Manually align to the next 8-byte boundary
+    // NOTE: GCC over-aligns to a full page when using .align directives on ARM (last tested on GCC 11.2)
+    "nop \n"
+    "0: \n"
+    ".quad 0, 0, 0, 0 \n" // TrampolineInstanceInfo
+  );
+#else
+#error Unsupported host architecture
+#endif
+}
+
+extern char __start_HostToGuestTrampolineTemplate[];
+extern char __stop_HostToGuestTrampolineTemplate[];
 
 namespace FEXCore {
     struct ExportEntry { uint8_t *sha256; ThunkedFunction* Fn; };
 
+    struct TrampolineInstanceInfo {
+      uintptr_t HostPacker;
+      uintptr_t CallCallback;
+      uintptr_t GuestUnpacker;
+      uintptr_t GuestTarget;
+    };
+
+    struct GuestcallInfo {
+      uintptr_t GuestUnpacker;
+      uintptr_t GuestTarget;
+
+      bool operator==(const GuestcallInfo&) const noexcept = default;
+    };
+
+    struct GuestcallInfoHash {
+      size_t operator()(const GuestcallInfo& x) const noexcept {
+        // Hash only the target address, which is generally unique.
+        // For the unlikely case of a hash collision, std::unordered_map still picks the correct bucket entry.
+        return std::hash<uintptr_t>{}(x.GuestTarget);
+      }
+    };
+
+    // Bits in a SHA256 sum are already randomly distributed, so truncation yields a suitable hash function
+    struct TruncatingSHA256Hash {
+      size_t operator()(const FEXCore::IR::SHA256Sum& SHA256Sum) const noexcept {
+        return (const size_t&)SHA256Sum;
+      }
+    };
+
     class ThunkHandler_impl final: public ThunkHandler {
         std::shared_mutex ThunksMutex;
 
-        std::map<IR::SHA256Sum, ThunkedFunction*> Thunks = {
+        std::unordered_map<IR::SHA256Sum, ThunkedFunction*, TruncatingSHA256Hash> Thunks = {
             {
                 // sha256(fex:loadlib)
-                { 0x27, 0x7e, 0xb7, 0x69, 0x5b, 0xe9, 0xab, 0x12, 0x6e, 0xf7, 0x85, 0x9d, 0x4b, 0xc9, 0xa2, 0x44, 0x46, 0xcf, 0xbd, 0xb5, 0x87, 0x43, 0xef, 0x28, 0xa2, 0x65, 0xba, 0xfc, 0x89, 0x0f, 0x77, 0x80},
+                { 0x27, 0x7e, 0xb7, 0x69, 0x5b, 0xe9, 0xab, 0x12, 0x6e, 0xf7, 0x85, 0x9d, 0x4b, 0xc9, 0xa2, 0x44, 0x46, 0xcf, 0xbd, 0xb5, 0x87, 0x43, 0xef, 0x28, 0xa2, 0x65, 0xba, 0xfc, 0x89, 0x0f, 0x77, 0x80 },
                 &LoadLib
+            },
+            {
+                // sha256(fex:is_lib_loaded)
+                { 0xee, 0x57, 0xba, 0x0c, 0x5f, 0x6e, 0xef, 0x2a, 0x8c, 0xb5, 0x19, 0x81, 0xc9, 0x23, 0xe6, 0x51, 0xae, 0x65, 0x02, 0x8f, 0x2b, 0x5d, 0x59, 0x90, 0x6a, 0x7e, 0xe2, 0xe7, 0x1c, 0x33, 0x8a, 0xff },
+                &IsLibLoaded
+            },
+            {
+                // sha256(fex:link_address_to_function)
+                { 0xe6, 0xa8, 0xec, 0x1c, 0x7b, 0x74, 0x35, 0x27, 0xe9, 0x4f, 0x5b, 0x6e, 0x2d, 0xc9, 0xa0, 0x27, 0xd6, 0x1f, 0x2b, 0x87, 0x8f, 0x2d, 0x35, 0x50, 0xea, 0x16, 0xb8, 0xc4, 0x5e, 0x42, 0xfd, 0x77 },
+                &LinkAddressToGuestFunction
+            },
+            {
+                // sha256(fex:make_host_trampoline_for_guest_function)
+                { 0x1e, 0x51, 0x6b, 0x07, 0x39, 0xeb, 0x50, 0x59, 0xb3, 0xf3, 0x4f, 0xca, 0xdd, 0x58, 0x37, 0xe9, 0xf0, 0x30, 0xe5, 0x89, 0x81, 0xc7, 0x14, 0xfb, 0x24, 0xf9, 0xba, 0xe7, 0x0e, 0x00, 0x1e, 0x86 },
+                &MakeHostTrampolineForGuestFunction
             }
         };
 
+        // Can't be a string_view. We need to keep a copy of the library name in-case string_view pointer goes away.
+        // Ideally we track when a library has been unloaded and remove it from this set before the memory backing goes away.
+        std::set<std::string> Libs;
+
+        std::unordered_map<GuestcallInfo, uintptr_t, GuestcallInfoHash> GuestcallToHostTrampoline;
+
+        uint8_t *HostTrampolineInstanceDataPtr;
+        size_t HostTrampolineInstanceDataAvailable = 0;
+
+
         /*
             Set arg0/1 to arg regs, use CTX::HandleCallback to handle the callback
         */
@@ -56,13 +140,160 @@ namespace FEXCore {
           Thread->CTX->HandleCallback(Thread, (uintptr_t)callback);
         }
 
+        /**
+         * Instructs the Core to redirect calls to functions at the given
+         * address to another function. The original callee address is passed
+         * to the target function through an implicit argument stored in r11.
+         *
+         * The primary use case of this is ensuring that host function pointers
+         * returned from thunked APIs can safely be called by the guest.
+         */
+        static void LinkAddressToGuestFunction(void* argsv) {
+            struct args_t {
+                uintptr_t original_callee;
+                uintptr_t target_addr;     // Guest function to call when branching to original_callee
+            };
+
+            auto args = reinterpret_cast<args_t*>(argsv);
+            auto CTX = Thread->CTX;
+
+            LOGMAN_THROW_A_FMT(args->original_callee, "Tried to link null pointer address to guest function");
+            LOGMAN_THROW_A_FMT(args->target_addr, "Tried to link address to null pointer guest function");
+            if (!CTX->Config.Is64BitMode) {
+                LOGMAN_THROW_A_FMT((args->original_callee >> 32) == 0, "Tried to link 64-bit address in 32-bit mode");
+                LOGMAN_THROW_A_FMT((args->target_addr >> 32) == 0, "Tried to link 64-bit address in 32-bit mode");
+            }
+
+            LogMan::Msg::DFmt("Thunks: Adding guest trampoline from address {:#x} to guest function {:#x}",
+                              args->original_callee, args->target_addr);
+
+            auto Result = Thread->CTX->AddCustomIREntrypoint(
+                    args->original_callee,
+                    [CTX, GuestThunkEntrypoint = args->target_addr](uintptr_t Entrypoint, FEXCore::IR::IREmitter *emit) {
+                        auto IRHeader = emit->_IRHeader(emit->Invalid(), 0);
+                        auto Block = emit->CreateCodeNode();
+                        IRHeader.first->Blocks = emit->WrapNode(Block);
+                        emit->SetCurrentCodeBlock(Block);
+
+                        const uint8_t GPRSize = CTX->GetGPRSize();
+
+                        emit->_StoreContext(GPRSize, IR::GPRClass, emit->_Constant(Entrypoint), offsetof(Core::CPUState, gregs[X86State::REG_R11]));
+                        emit->_ExitFunction(emit->_Constant(GuestThunkEntrypoint));
+                    }, CTX->ThunkHandler.get(), (void*)args->target_addr);
+
+            if (!Result) {
+                if (Result.Creator != CTX->ThunkHandler.get()) {
+                    ERROR_AND_DIE_FMT("Input address for LinkAddressToGuestFunction is already linked by another module");
+                }
+                if (Result.Data != (void*)args->target_addr) {
+                    // NOTE: This may happen in Vulkan thunks if the Vulkan driver resolves two different symbols
+                    //       to the same function (e.g. vkGetPhysicalDeviceFeatures2/vkGetPhysicalDeviceFeatures2KHR)
+                    LogMan::Msg::EFmt("Input address for LinkAddressToGuestFunction is already linked elsewhere");
+                }
+            }
+        }
+
+        /**
+         * Generates a host-callable trampoline to call guest functions via the host ABI.
+         *
+         * This trampoline uses the same calling convention as the given HostPacker. Trampolines
+         * are cached, so it's safe to call this function repeatedly on the same arguments without
+         * leaking memory.
+         *
+         * Invoking the returned trampoline has the effect of:
+         * - packing the arguments (using the HostPacker identified by its SHA256)
+         * - performing a host->guest transition
+         * - unpacking the arguments via GuestUnpacker
+         * - calling the function at GuestTarget
+         *
+         * The primary use case of this is ensuring that guest function pointers ("callbacks")
+         * passed to thunked APIs can safely be called by the native host library.
+         */
+        static void MakeHostTrampolineForGuestFunction(void* ArgsRV) {
+          struct ArgsRV_t {
+              IR::SHA256Sum *HostPackerSha256;
+              uintptr_t GuestUnpacker;
+              uintptr_t GuestTarget;
+              uintptr_t rv; // Pointer to host trampoline + TrampolineInstanceInfo
+          } *args = reinterpret_cast<ArgsRV_t*>(ArgsRV);
+
+          LOGMAN_THROW_A_FMT(args->GuestTarget, "Tried to create host-trampoline to null pointer guest function");
+
+          const auto CTX = Thread->CTX;
+          const auto ThunkHandler = reinterpret_cast<ThunkHandler_impl *>(CTX->ThunkHandler.get());
+
+          const GuestcallInfo gci = { args->GuestUnpacker, args->GuestTarget };
+
+          // Try first with shared_lock
+          {
+            std::shared_lock lk(ThunkHandler->ThunksMutex);
+
+            auto found = ThunkHandler->GuestcallToHostTrampoline.find(gci);
+            if (found != ThunkHandler->GuestcallToHostTrampoline.end()) {
+              args->rv = found->second;
+              return;
+            }
+          }
+
+          std::lock_guard lk(ThunkHandler->ThunksMutex);
+
+          // Retry lookup with full lock before making a new trampoline to avoid double trampolines
+          {
+            auto found = ThunkHandler->GuestcallToHostTrampoline.find(gci);
+            if (found != ThunkHandler->GuestcallToHostTrampoline.end()) {
+              args->rv = found->second;
+              return;
+            }
+          }
+
+          // No entry found => create new trampoline
+          auto HostPackerEntry = ThunkHandler->Thunks.find(*args->HostPackerSha256);
+          if (HostPackerEntry == ThunkHandler->Thunks.end()) {
+            ERROR_AND_DIE_FMT("Unknown host packing function for callback");
+          }
+
+          LogMan::Msg::DFmt("Thunks: Adding host trampoline for guest function {:#x}",
+                            args->GuestTarget);
+
+          const auto Length = __stop_HostToGuestTrampolineTemplate - __start_HostToGuestTrampolineTemplate;
+          const auto InstanceInfoOffset = Length - sizeof(TrampolineInstanceInfo);
+
+          if (ThunkHandler->HostTrampolineInstanceDataAvailable < Length) {
+            const auto allocation_step = 16 * 1024;
+            ThunkHandler->HostTrampolineInstanceDataAvailable = allocation_step;
+            ThunkHandler->HostTrampolineInstanceDataPtr = (uint8_t *)mmap(
+                0, ThunkHandler->HostTrampolineInstanceDataAvailable,
+                PROT_READ | PROT_WRITE | PROT_EXEC,
+                MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+
+            LOGMAN_THROW_A_FMT(ThunkHandler->HostTrampolineInstanceDataPtr != MAP_FAILED, "Failed to mmap HostTrampolineInstanceDataPtr");
+          }
+
+          const TrampolineInstanceInfo NewTrampolineInfo {
+            .HostPacker = reinterpret_cast<uintptr_t>(HostPackerEntry->second),
+            .CallCallback = (uintptr_t)&CallCallback,
+            .GuestUnpacker = args->GuestUnpacker,
+            .GuestTarget = args->GuestTarget
+          };
+
+          uint8_t* const HostTrampoline = ThunkHandler->HostTrampolineInstanceDataPtr;
+          ThunkHandler->HostTrampolineInstanceDataAvailable -= Length;
+          ThunkHandler->HostTrampolineInstanceDataPtr += Length;
+
+          memcpy(HostTrampoline, (void*)&HostToGuestTrampolineTemplate, Length);
+          memcpy(HostTrampoline + InstanceInfoOffset, &NewTrampolineInfo, sizeof(NewTrampolineInfo));
+
+          args->rv = reinterpret_cast<uintptr_t>(HostTrampoline);
+
+          ThunkHandler->GuestcallToHostTrampoline[gci] = args->rv;
+        }
+
         static void LoadLib(void *ArgsV) {
             auto CTX = Thread->CTX;
 
             auto Args = reinterpret_cast<LoadlibArgs*>(ArgsV);
 
             auto Name = Args->Name;
-            auto CallbackThunks = Args->CallbackThunks;
 
             auto SOName = CTX->Config.ThunkHostLibsPath() + "/" + (const char*)Name + "-host.so";
 
@@ -75,13 +306,13 @@ namespace FEXCore {
 
             const auto InitSym = std::string("fexthunks_exports_") + Name;
 
-            ExportEntry* (*InitFN)(void *, uintptr_t);
+            ExportEntry* (*InitFN)();
             (void*&)InitFN = dlsym(Handle, InitSym.c_str());
             if (!InitFN) {
                 ERROR_AND_DIE_FMT("LoadLib: Failed to find export {}", InitSym);
             }
 
-            auto Exports = InitFN((void*)&CallCallback, CallbackThunks);
+            auto Exports = InitFN();
             if (!Exports) {
                 ERROR_AND_DIE_FMT("LoadLib: Failed to initialize thunk library {}. "
                                   "Check if the corresponding host library is installed "
@@ -91,7 +322,9 @@ namespace FEXCore {
             auto That = reinterpret_cast<ThunkHandler_impl*>(CTX->ThunkHandler.get());
 
             {
-                std::unique_lock lk(That->ThunksMutex);
+                std::lock_guard lk(That->ThunksMutex);
+
+                That->Libs.insert(Name);
 
                 int i;
                 for (i = 0; Exports[i].sha256; i++) {
@@ -102,6 +335,23 @@ namespace FEXCore {
             }
         }
 
+        static void IsLibLoaded(void* ArgsRV) {
+            struct ArgsRV_t {
+                const char *Name;
+                bool rv;
+            };
+
+            auto &[Name, rv] = *reinterpret_cast<ArgsRV_t*>(ArgsRV);
+
+            auto CTX = Thread->CTX;
+            auto That = reinterpret_cast<ThunkHandler_impl*>(CTX->ThunkHandler.get());
+
+            {
+                std::shared_lock lk(That->ThunksMutex);
+                rv = That->Libs.contains(Name);
+            }
+        }
+
         public:
 
         ThunkedFunction* LookupThunk(const IR::SHA256Sum &sha256) {
@@ -120,12 +370,6 @@ namespace FEXCore {
         void RegisterTLSState(FEXCore::Core::InternalThreadState *Thread) {
             ::Thread = Thread;
         }
-
-        ThunkHandler_impl() {
-        }
-
-        ~ThunkHandler_impl() {
-        }
     };
 
     ThunkHandler* ThunkHandler::Create() {

External/FEXCore/Source/Interface/IR/AOTIR.cpp

@@ -4,6 +4,9 @@
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/IR/RegisterAllocationData.h>
 #include <FEXCore/Utils/Allocator.h>
+#include <FEXCore/HLE/SyscallHandler.h>
+#include <Interface/Core/LookupCache.h>
+#include <Interface/GDBJIT/GDBJIT.h>
 
 #include <cstddef>
 #include <cstdint>
@@ -15,6 +18,7 @@
 #include <unistd.h>
 #include <xxhash.h>
 
+
 namespace FEXCore::IR {
   AOTIRInlineEntry *AOTIRInlineIndex::GetInlineEntry(uint64_t DataOffset) {
     uintptr_t This = (uintptr_t)this;
@@ -77,7 +81,7 @@ namespace FEXCore::IR {
       return true;
   }
 
-  bool LoadAOTIRCache(AOTCacheType *AOTIRCache, int streamfd) {
+  static bool LoadAOTIRCache(AOTIRCacheEntry *Entry, int streamfd) {
     uint64_t tag;
 
     if (!readAll(streamfd, (char*)&tag, sizeof(tag)) || tag != FEXCore::IR::AOTIR_COOKIE)
@@ -99,6 +103,10 @@ namespace FEXCore::IR {
     if (!readAll(streamfd,  (char*)&Module[0], Module.size()))
       return false;
 
+    if (Entry->FileId != Module) {
+      return false;
+    }
+
     lseek(streamfd, -sizeof(ModSize) - ModSize - sizeof(IndexSize), SEEK_END);
 
     if (!readAll(streamfd,  (char*)&IndexSize, sizeof(IndexSize)))
@@ -119,19 +127,16 @@ namespace FEXCore::IR {
 
     auto Array = (AOTIRInlineIndex *)((char*)FilePtr + IndexOffset);
 
-    AOTIRCache->insert({Module, {Array, FilePtr, Size}});
+    LOGMAN_THROW_A_FMT(Entry->Array == nullptr && Entry->FilePtr == nullptr, "Entry must not be initialized here");
+    Entry->Array = Array;
+    Entry->FilePtr = FilePtr;
+    Entry->Size = Size;
 
     LogMan::Msg::DFmt("AOTIR: Module {} has {} functions", Module, Array->Count);
 
     return true;
   }
 
-  AOTIRCaptureCache::~AOTIRCaptureCache() {
-    for (auto &Mod: AOTIRCache) {
-      FEXCore::Allocator::munmap(Mod.second.mapping, Mod.second.size);
-    }
-  }
-
   void AOTIRCaptureCache::FinalizeAOTIRCache() {
     AOTIRCaptureCacheWriteoutQueue_Flush();
 
@@ -230,39 +235,27 @@ namespace FEXCore::IR {
 
   void AOTIRCaptureCache::WriteFilesWithCode(std::function<void(const std::string& fileid, const std::string& filename)> Writer) {
     std::shared_lock lk(AOTIRCacheLock);
-    for( const auto &File: FilesWithCode) {
-      Writer(File.first, File.second);
+    for( const auto &Entry: AOTIRCache) {
+      if (Entry.second.ContainsCode) {
+        Writer(Entry.second.FileId, Entry.second.Filename);
+      }
     }
   }
 
   AOTIRCaptureCache::PreGenerateIRFetchResult AOTIRCaptureCache::PreGenerateIRFetch(uint64_t GuestRIP, FEXCore::IR::IRListView *IRList) {
-    {
-      std::shared_lock lk(AOTIRCacheLock);
-      auto file = AddrToFile.lower_bound(GuestRIP);
-      if (file != AddrToFile.begin()) {
-        --file;
-        if (!file->second.ContainsCode) {
-          file->second.ContainsCode = true;
-          FilesWithCode[file->second.fileid] = file->second.filename;
-        }
-      }
-    }
+    auto AOTIRCacheEntry = CTX->SyscallHandler->LookupAOTIRCacheEntry(GuestRIP);
 
     PreGenerateIRFetchResult Result{};
-    if (IRList == nullptr && CTX->Config.AOTIRLoad()) {
-      std::shared_lock lk(AOTIRCacheLock);
-      auto file = AddrToFile.lower_bound(GuestRIP);
-      if (file != AddrToFile.begin()) {
-        --file;
-        auto Mod = (FEXCore::IR::AOTIRInlineIndex*)file->second.CachedFileEntry;
 
-        if (Mod == nullptr) {
-          file->second.CachedFileEntry = Mod = AOTIRCache[file->second.fileid].Array;
-        }
+    if (AOTIRCacheEntry.Entry) {
+      AOTIRCacheEntry.Entry->ContainsCode = true;
+
+      if (IRList == nullptr && CTX->Config.AOTIRLoad()) {
+        auto Mod = AOTIRCacheEntry.Entry->Array;
 
         if (Mod != nullptr)
         {
-          auto AOTEntry = Mod->Find(GuestRIP - file->second.Start + file->second.Offset);
+          auto AOTEntry = Mod->Find(GuestRIP - AOTIRCacheEntry.VAFileStart);
 
           if (AOTEntry) {
             // verify hash
@@ -272,7 +265,7 @@ namespace FEXCore::IR {
               Result.IRList = AOTEntry->GetIRData();
               //LogMan::Msg::DFmt("using {} + {:x} -> {:x}\n", file->second.fileid, AOTEntry->first, GuestRIP);
 
-              Result.RAData = AOTEntry->GetRAData();;
+              Result.RAData = AOTEntry->GetRAData()->CreateCopy();
               Result.DebugData = new FEXCore::Core::DebugData();
               Result.StartAddr = MappedStart;
               Result.Length = AOTEntry->GuestLength;
@@ -296,21 +289,23 @@ namespace FEXCore::IR {
     uint64_t GuestRIP,
     uint64_t StartAddr,
     uint64_t Length,
-    FEXCore::IR::RegisterAllocationData *RAData,
+    FEXCore::IR::RegisterAllocationData::UniquePtr RAData,
     FEXCore::IR::IRListView *IRList,
     FEXCore::Core::DebugData *DebugData,
-    bool GeneratedIR,
-    bool DecrementRefCount) {
+    bool GeneratedIR) {
+
     // Both generated ir and LibraryJITName need a named region lookup
-    if (GeneratedIR || CTX->Config.LibraryJITNaming()) {
-      std::shared_lock lk(AOTIRCacheLock);
+    if (GeneratedIR || CTX->Config.LibraryJITNaming() || CTX->Config.GDBSymbols()) {
 
-      auto file = FindAddrForFile(StartAddr, Length);
+      auto AOTIRCacheEntry = CTX->SyscallHandler->LookupAOTIRCacheEntry(GuestRIP);
 
-      // Only go down this path if we actually found a library region
-      if (file != AddrToFile.end()) {
+      if (AOTIRCacheEntry.Entry) {
         if (DebugData && CTX->Config.LibraryJITNaming()) {
-          CTX->Symbols.RegisterNamedRegion(CodePtr, DebugData->HostCodeSize, file->second.filename);
+          CTX->Symbols.RegisterNamedRegion(CodePtr, DebugData->HostCodeSize, AOTIRCacheEntry.Entry->Filename);
+        }
+
+        if (CTX->Config.GDBSymbols()) {
+          GDBJITRegister(AOTIRCacheEntry.Entry, AOTIRCacheEntry.VAFileStart, GuestRIP, (uintptr_t)CodePtr, DebugData);
         }
 
         // Add to AOT cache if aot generation is enabled
@@ -319,29 +314,35 @@ namespace FEXCore::IR {
 
           auto hash = XXH3_64bits((void*)StartAddr, Length);
 
-          auto LocalRIP = GuestRIP - file->second.Start + file->second.Offset;
-          auto LocalStartAddr = StartAddr - file->second.Start + file->second.Offset;
-          auto fileid = file->second.fileid;
-          AOTIRCaptureCacheWriteoutQueue_Append([this, LocalRIP, LocalStartAddr, Length, hash, IRList, RAData, fileid]() {
-            auto *AotFile = &AOTIRCaptureCacheMap[fileid];
+          auto LocalRIP = GuestRIP - AOTIRCacheEntry.VAFileStart;
+          auto LocalStartAddr = StartAddr - AOTIRCacheEntry.VAFileStart;
+          auto FileId = AOTIRCacheEntry.Entry->FileId;
+          // The underlying pointer and the unique_ptr deleter for RAData must
+          // be marshalled separately to the lambda below. Otherwise, the
+          // lambda can't be used as an std::function due to being non-copyable
+          auto RADataCopy = RAData->CreateCopy();
+          auto RADataCopyDeleter = RADataCopy.get_deleter();
+          auto IRListCopy = IRList->CreateCopy();
+          AOTIRCaptureCacheWriteoutQueue_Append([this, LocalRIP, LocalStartAddr, Length, hash, IRListCopy, RADataCopy=RADataCopy.release(), RADataCopyDeleter, FileId]() {
+
+            // It is guaranteed via AOTIRCaptureCacheWriteoutLock and AOTIRCaptureCacheWriteoutFlusing that this will not run concurrently
+            // Memory coherency is guaranteed via AOTIRCaptureCacheWriteoutLock
+            
+            auto *AotFile = &AOTIRCaptureCacheMap[FileId];
 
             if (!AotFile->Stream) {
-              AotFile->Stream = AOTIRWriter(fileid);
+              AotFile->Stream = AOTIRWriter(FileId);
               uint64_t tag = FEXCore::IR::AOTIR_COOKIE;
               AotFile->Stream->write((char*)&tag, sizeof(tag));
             }
-            AotFile->AppendAOTIRCaptureCache(LocalRIP, LocalStartAddr, Length, hash, IRList, RAData);
+            AotFile->AppendAOTIRCaptureCache(LocalRIP, LocalStartAddr, Length, hash, IRListCopy, RADataCopy);
+            RADataCopyDeleter(RADataCopy);
+            delete IRListCopy;
           });
 
           if (CTX->Config.AOTIRGenerate()) {
             // cleanup memory and early exit here -- we're not running the application
-
-            if (DecrementRefCount) {
-              --Thread->CompileBlockReentrantRefCount;
-            }
-
             Thread->CPUBackend->ClearCache();
-
             return true;
           }
         }
@@ -349,30 +350,25 @@ namespace FEXCore::IR {
 
       // Insert to caches if we generated IR
       if (GeneratedIR) {
-        // Add to thread local ir cache
-        Core::LocalIREntry Entry = {StartAddr, Length, decltype(Entry.IR)(IRList), decltype(Entry.RAData)(RAData), decltype(Entry.DebugData)(DebugData)};
-        Thread->LocalIRCache.insert({GuestRIP, std::move(Entry)});
+        if (CTX->GetGdbServerStatus()) {
+          // Add to thread local ir cache
+          Core::LocalIREntry Entry = {StartAddr, Length, decltype(Entry.IR)(IRList), std::move(RAData), decltype(Entry.DebugData)(DebugData)};
+          
+          std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
+          Thread->DebugStore.insert({GuestRIP, std::move(Entry)});
+        }
+        else {
+          // If the IR doesn't need to be retained then we can just delete it now
+          delete DebugData;
+          if (IRList->IsCopy()) delete IRList;
+        }
       }
     }
 
     return false;
   }
 
-  AOTIRCaptureCache::AddrToFileMapType::iterator AOTIRCaptureCache::FindAddrForFile(uint64_t Entry, uint64_t Length) {
-    // Thread safety here! We are returning an iterator to the map object
-    // This needs the AOTIRCacheLock locked prior to coming in to the function
-    auto file = AddrToFile.lower_bound(Entry);
-    if (file != AddrToFile.begin()) {
-      --file;
-      if (file->second.Start <= Entry && (file->second.Start + file->second.Len) >= (Entry + Length)) {
-        return file;
-      }
-    }
-    return AddrToFile.end();
-  }
-
-  void AOTIRCaptureCache::AddNamedRegion(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename) {
-    // TODO: Support overlapping maps and region splitting
+  AOTIRCacheEntry *AOTIRCaptureCache::LoadAOTIRCacheEntry(const std::string &filename) {
     auto base_filename = std::filesystem::path(filename).filename().string();
 
     if (!base_filename.empty()) {
@@ -388,21 +384,32 @@ namespace FEXCore::IR {
 
       std::unique_lock lk(AOTIRCacheLock);
 
-      AddrToFile.insert({ Base, { Base, Size, Offset, fileid, filename, nullptr, false} });
+      auto Inserted = AOTIRCache.insert({fileid, AOTIRCacheEntry { .FileId = fileid, .Filename = filename }});
+      auto Entry = &(Inserted.first->second);
 
-      if (CTX->Config.AOTIRLoad && !AOTIRCache.contains(fileid) && AOTIRLoader) {
+      LOGMAN_THROW_A_FMT(Entry->Array == nullptr, "Duplicate LoadAOTIRCacheEntry");
+
+      if (CTX->Config.AOTIRLoad && AOTIRLoader) {
         auto streamfd = AOTIRLoader(fileid);
         if (streamfd != -1) {
-          FEXCore::IR::LoadAOTIRCache(&AOTIRCache, streamfd);
+          FEXCore::IR::LoadAOTIRCache(Entry, streamfd);
           close(streamfd);
         }
       }
+      return Entry;
     }
+
+    return nullptr;
   }
 
-  void AOTIRCaptureCache::RemoveNamedRegion(uintptr_t Base, uintptr_t Size) {
-    std::unique_lock lk(AOTIRCacheLock);
-    // TODO: Support partial removing
-    AddrToFile.erase(Base);
+  void AOTIRCaptureCache::UnloadAOTIRCacheEntry(AOTIRCacheEntry *Entry) {
+    LOGMAN_THROW_A_FMT(Entry != nullptr, "Removing not existing entry");
+
+    if (Entry->Array) {
+      FEXCore::Allocator::munmap(Entry->FilePtr, Entry->Size);
+      Entry->Array = nullptr;
+      Entry->FilePtr = nullptr;
+      Entry->Size = 0;
+    }
   }
 }

External/FEXCore/Source/Interface/IR/AOTIR.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include "FEXCore/IR/RegisterAllocationData.h"
 #include <FEXCore/Config/Config.h>
 
 #include <atomic>
@@ -11,6 +12,7 @@
 #include <unordered_map>
 #include <shared_mutex>
 #include <queue>
+#include <FEXCore/HLE/SourcecodeResolver.h>
 
 namespace FEXCore::Core {
 struct DebugData;
@@ -72,18 +74,20 @@ namespace FEXCore::IR {
 
   struct AOTIRCacheEntry {
     AOTIRInlineIndex *Array;
-    void *mapping;
-    size_t size;
+    void *FilePtr;
+    size_t Size;
+    std::unique_ptr<FEXCore::HLE::SourcecodeMap> SourcecodeMap;
+    std::string FileId;
+    std::string Filename;
+    bool ContainsCode;
   };
 
   using AOTCacheType = std::unordered_map<std::string, FEXCore::IR::AOTIRCacheEntry>;
-  bool LoadAOTIRCache(AOTCacheType *AOTIRCache, int streamfd);
 
   class AOTIRCaptureCache final {
     public:
 
       AOTIRCaptureCache(FEXCore::Context::Context *ctx) : CTX {ctx} {}
-      ~AOTIRCaptureCache();
 
       void FinalizeAOTIRCache();
       void AOTIRCaptureCacheWriteoutQueue_Flush();
@@ -92,7 +96,7 @@ namespace FEXCore::IR {
 
       struct PreGenerateIRFetchResult {
         FEXCore::IR::IRListView *IRList {};
-        FEXCore::IR::RegisterAllocationData *RAData {};
+        FEXCore::IR::RegisterAllocationData::UniquePtr RAData {};
         FEXCore::Core::DebugData *DebugData {};
         uint64_t StartAddr {};
         uint64_t Length {};
@@ -105,14 +109,13 @@ namespace FEXCore::IR {
         uint64_t GuestRIP,
         uint64_t StartAddr,
         uint64_t Length,
-        FEXCore::IR::RegisterAllocationData *RAData,
+        FEXCore::IR::RegisterAllocationData::UniquePtr RAData,
         FEXCore::IR::IRListView *IRList,
         FEXCore::Core::DebugData *DebugData,
-        bool GeneratedIR,
-        bool DecrementRefCount);
+        bool GeneratedIR);
 
-      void AddNamedRegion(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename);
-      void RemoveNamedRegion(uintptr_t Base, uintptr_t Size);
+      AOTIRCacheEntry *LoadAOTIRCacheEntry(const std::string &filename);
+      void UnloadAOTIRCacheEntry(AOTIRCacheEntry *Entry);
 
       // Callbacks
       void SetAOTIRLoader(std::function<int(const std::string&)> CacheReader) {
@@ -136,27 +139,11 @@ namespace FEXCore::IR {
 
       std::queue<std::function<void()>> AOTIRCaptureCacheWriteoutQueue;
 
-      std::map<std::string, std::string> FilesWithCode;
-
-      struct AddrToFileEntry {
-        uint64_t Start;
-        uint64_t Len;
-        uint64_t Offset;
-        std::string fileid;
-        std::string filename;
-        void *CachedFileEntry;
-        bool ContainsCode;
-      };
-
-      using AddrToFileMapType = std::map<uint64_t, AddrToFileEntry>;
-      AddrToFileMapType AddrToFile;
       FEXCore::IR::AOTCacheType AOTIRCache;
 
       std::function<int(const std::string&)> AOTIRLoader;
       std::function<std::unique_ptr<std::ofstream>(const std::string&)> AOTIRWriter;
       std::function<void(const std::string&)> AOTIRRenamer;
       std::unordered_map<std::string, FEXCore::IR::AOTIRCaptureCacheEntry> AOTIRCaptureCacheMap;
-
-      AddrToFileMapType::iterator FindAddrForFile(uint64_t Entry, uint64_t Length);
   };
 }

External/FEXCore/Source/Interface/IR/IR.json

@@ -181,13 +181,18 @@
         "RAOverride": "0"
       },
 
+      "GuestOpcode u32:$GuestEntryOffset": {
+        "Desc": ["Marks the beginning of a guest opcode"],
+        "HasSideEffects": true
+      },
+
       "GPR = ValidateCode u64:$CodeOriginalLow, u64:$CodeOriginalhigh, i64:$Offset, u8:$CodeLength": {
         "HasSideEffects": true,
         "HasDest": true,
         "DestSize": "8"
       },
 
-      "RemoveCodeEntry": {
+      "RemoveThreadCodeEntry": {
         "HasSideEffects": true
       },
 
@@ -231,6 +236,11 @@
                 ],
         "DestSize": "16",
         "NumElements": "2"
+      },
+      "Yield": {
+        "HasSideEffects": true,
+        "Desc": ["This is a hint instruction that the CPU is likely to do a spin so it might want to pause to help out SMP",
+                 "Can be implemented as a NOP if necessary"]
       }
     },
     "Branch": {
@@ -289,12 +299,6 @@
                 ],
         "DestSize": "16",
         "NumElements": "2"
-      },
-      "GuestCallDirect u64:$RIP, u64:$NextRIP": {
-        "HasSideEffects": true
-      },
-      "GuestCallIndirect GPR:$RIP, u64:$NextRIP": {
-        "HasSideEffects": true
       }
     },
     "Moves": {
@@ -938,7 +942,7 @@
       },
 
       "FPR = VBitcast u8:#RegisterSize, u8:#ElementSize, FPR:$Source": {
-        "Dest": ["Workaround for issue with LLVM breaking when loading scalar elements to vectors"],
+        "Desc": ["Workaround for issue with LLVM breaking when loading scalar elements to vectors"],
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
@@ -1272,12 +1276,12 @@
         "NumElements": "RegisterSize / (ElementSize << 1)"
       },
       "FPR = VUMull2 u8:#RegisterSize, u8:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
-        "Dest": "Multiplies the high elements with size extension",
+        "Desc": "Multiplies the high elements with size extension",
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / (ElementSize << 1)"
       },
       "FPR = VSMull2 u8:#RegisterSize, u8:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
-        "Dest": "Multiplies the high elements with size extension",
+        "Desc": "Multiplies the high elements with size extension",
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / (ElementSize << 1)"
       },
@@ -1449,35 +1453,74 @@
     },
     "Crypto": {
       "FPR = VAESImc FPR:$Vector": {
-        "Dest": "Does a stage of the inverse mix column transformation",
+        "Desc": "Does a stage of the inverse mix column transformation",
         "DestSize": "16"
       },
       "FPR = VAESEnc FPR:$State, FPR:$Key": {
-        "Dest": "Does a step of AES encryption",
+        "Desc": "Does a step of AES encryption",
         "DestSize": "16"
       },
       "FPR = VAESEncLast FPR:$State, FPR:$Key": {
-        "Dest": "Does the last step of AES encryption",
+        "Desc": "Does the last step of AES encryption",
         "DestSize": "16"
       },
       "FPR = VAESDec FPR:$State, FPR:$Key": {
-        "Dest": "Does a step of AES decryption",
+        "Desc": "Does a step of AES decryption",
         "DestSize": "16"
       },
       "FPR = VAESDecLast FPR:$State, FPR:$Key": {
-        "Dest": "Does the last step of AES decryption",
+        "Desc": "Does the last step of AES decryption",
         "DestSize": "16"
       },
       "FPR = VAESKeyGenAssist FPR:$Src, u8:$RCON": {
-        "Dest": "Assists in key generation",
+        "Desc": "Assists in key generation",
         "DestSize": "16"
       },
       "GPR = CRC32 GPR:$Src1, GPR:$Src2, u8:$SrcSize": {
         "Desc": ["CRC32 using polynomial 0x1EDC6F41"
                 ],
         "DestSize": "std::max<uint8_t>(4, GetOpSize(_Src1))"
+      },
+      "FPR = PCLMUL FPR:$Src1, FPR:$Src2, u8:$Selector": {
+        "Desc": [
+          "Performs carryless multiplication of 64-bit elements depending on the selector.",
+          "Selector = 0b00000000: Uses low 64-bit elements from both input vectors",
+          "Selector = 0b00000001: Uses high 64-bit element from Src1 and low 64-bit element from Src2",
+          "Selector = 0b00010000: Uses low 64-bit element from Src1 and high 64-bit element from Src2",
+          "Selector = 0b00010001: Uses high 64-bit elements from both input vectors"
+        ],
+        "DestSize": "16"
       }
     },
+    "F64": {
+        "FPR = F64ATAN FPR:$Src1, FPR:$Src2": {
+            "DestSize": "8"
+        },
+        "FPR = F64FPREM FPR:$Src1, FPR:$Src2": {
+            "DestSize": "8"
+        },
+        "FPR = F64FPREM1 FPR:$Src1, FPR:$Src2": {
+            "DestSize": "8"
+        },
+        "FPR = F64SCALE FPR:$Src1, FPR:$Src2": {
+            "DestSize": "8"
+        },
+        "FPR = F64F2XM1 FPR:$Src": {
+            "DestSize": "8"
+        },
+        "FPR = F64FYL2X FPR:$Src, FPR:$Src2": {
+            "DestSize": "8"
+        },
+        "FPR = F64TAN FPR:$Src": {
+            "DestSize": "8"
+        },
+        "FPR = F64SIN FPR:$Src": {
+            "DestSize": "8"
+        },
+        "FPR = F64COS FPR:$Src": {
+            "DestSize": "8"
+        }
+    },
     "F80": {
       "F80LoadFCW GPR:$Src": {
         "HasSideEffects": true

External/FEXCore/Source/Interface/IR/IREmitter.cpp

@@ -16,6 +16,27 @@ $end_info$
 #include <vector>
 
 namespace FEXCore::IR {
+
+bool IsFragmentExit(FEXCore::IR::IROps Op) {
+  switch (Op) {
+    case OP_EXITFUNCTION:
+    case OP_BREAK:
+      return true;
+    default:
+    return false;
+  }
+}
+
+bool IsBlockExit(FEXCore::IR::IROps Op) {
+  switch(Op) {
+    case OP_JUMP:
+    case OP_CONDJUMP:
+      return true;
+    default:
+      return IsFragmentExit(Op);
+  }
+}
+
 FEXCore::IR::RegisterClassType IREmitter::WalkFindRegClass(OrderedNode *Node) {
   auto Class = GetOpRegClass(Node);
   switch (Class) {

External/FEXCore/Source/Interface/IR/IRParser.cpp

@@ -5,6 +5,8 @@ tags: ir|parser
 $end_info$
 */
 
+#include "Common/StringUtils.h"
+
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/IR/IREmitter.h>
@@ -42,28 +44,6 @@ enum class DecodeFailure {
 
 };
 
-std::string ltrim(std::string String) {
-	size_t pos = std::string::npos;
-	if ((pos = String.find_first_not_of(" \t\n\r")) != std::string::npos) {
-		String.erase(0, pos);
-	}
-
-	return String;
-}
-
-std::string rtrim(std::string String) {
-	size_t pos = std::string::npos;
-	if ((pos = String.find_last_not_of(" \t\n\r")) != std::string::npos) {
-		String.erase(String.begin() + pos + 1, String.end());
-	}
-
-	return String;
-}
-
-std::string trim(std::string String) {
-  return rtrim(ltrim(String));
-}
-
 std::string DecodeErrorToString(DecodeFailure Failure) {
   switch (Failure) {
     case DecodeFailure::DECODE_OKAY: return "Okay";
@@ -295,7 +275,7 @@ class IRParser: public FEXCore::IR::IREmitter {
     if (Arg.at(0) != '%') return {DecodeFailure::DECODE_INVALIDCHAR, 0};
 
     // Strip off the type qualifier from the ssa value
-    std::string SSAName = trim(Arg);
+    std::string SSAName = FEXCore::StringUtils::Trim(Arg);
     const size_t ArgEnd = SSAName.find_first_of(' ');
 
     if (ArgEnd != std::string::npos) {
@@ -329,7 +309,8 @@ class IRParser: public FEXCore::IR::IREmitter {
   LineDefinition *CurrentDef{};
   std::unordered_map<std::string_view, FEXCore::IR::IROps> NameToOpMap;
 
-  IRParser(std::istream *text) {
+  IRParser(FEXCore::Utils::IntrusivePooledAllocator &ThreadAllocator, std::istream *text)
+    : IREmitter {ThreadAllocator} {
     InitializeNameMap();
 
     std::string TmpLine;
@@ -382,7 +363,7 @@ class IRParser: public FEXCore::IR::IREmitter {
       CurrentDef = &Def;
       Def.LineNumber = i;
 
-      Line = trim(Line);
+      Line = FEXCore::StringUtils::Trim(Line);
 
       // Skip empty lines
       if (Line.empty()) {
@@ -401,7 +382,7 @@ class IRParser: public FEXCore::IR::IREmitter {
         size_t DefinitionEnd = std::string::npos;
         if ((DefinitionEnd = Line.find_first_of('=', CurrentPos)) != std::string::npos) {
           Def.Definition = Line.substr(0, DefinitionEnd);
-          Def.Definition = trim(Def.Definition);
+          Def.Definition = FEXCore::StringUtils::Trim(Def.Definition);
           Def.HasDefinition = true;
           CurrentPos = DefinitionEnd + 1; // +1 to ensure we go past then assignment
         }
@@ -421,7 +402,7 @@ class IRParser: public FEXCore::IR::IREmitter {
           size_t SSAEnd = std::string::npos;
           if ((SSAEnd = Line.find_last_of(' ', DefinitionEnd)) != std::string::npos) {
             std::string Type = Line.substr(SSAEnd + 1, DefinitionEnd - SSAEnd - 1);
-            Type = trim(Type);
+            Type = FEXCore::StringUtils::Trim(Type);
 
             auto DefinitionSize = DecodeValue<FEXCore::IR::TypeDefinition>(Type);
             if (!CheckPrintError(Def, DefinitionSize.first)) {
@@ -430,7 +411,7 @@ class IRParser: public FEXCore::IR::IREmitter {
             Def.Size = DefinitionSize.second;
           }
 
-          Def.Definition = trim(Line.substr(1, std::min(DefinitionEnd, SSAEnd) - 1));
+          Def.Definition = FEXCore::StringUtils::Trim(Line.substr(1, std::min(DefinitionEnd, SSAEnd) - 1));
 
           CurrentPos = DefinitionEnd + 1;
         }
@@ -447,8 +428,8 @@ class IRParser: public FEXCore::IR::IREmitter {
         size_t NameEnd = std::string::npos;
         if ((NameEnd = Def.Definition.find_first_of(' ')) != std::string::npos) {
           std::string Type = Def.Definition.substr(NameEnd + 1);
-          Type = trim(Type);
-          Def.Definition = trim(Def.Definition.substr(0, NameEnd));
+          Type = FEXCore::StringUtils::Trim(Type);
+          Def.Definition = FEXCore::StringUtils::Trim(Def.Definition.substr(0, NameEnd));
 
           auto DefinitionSize = DecodeValue<FEXCore::IR::TypeDefinition>(Type);
           if (!CheckPrintError(Def, DefinitionSize.first)) return false;
@@ -465,11 +446,11 @@ class IRParser: public FEXCore::IR::IREmitter {
 
       // Let's get the IR op
       size_t OpNameEnd = std::string::npos;
-      std::string RemainingLine = trim(Line.substr(CurrentPos));
+      std::string RemainingLine = FEXCore::StringUtils::Trim(Line.substr(CurrentPos));
       CurrentPos = 0;
       if ((OpNameEnd = RemainingLine.find_first_of(" \t\n\r\0", CurrentPos)) != std::string::npos) {
         Def.IROp = RemainingLine.substr(CurrentPos, OpNameEnd);
-        Def.IROp = trim(Def.IROp);
+        Def.IROp = FEXCore::StringUtils::Trim(Def.IROp);
         Def.HasArgs = true;
         CurrentPos = OpNameEnd;
       }
@@ -486,7 +467,7 @@ class IRParser: public FEXCore::IR::IREmitter {
       }
 
       if (Def.HasArgs) {
-        RemainingLine = trim(RemainingLine.substr(CurrentPos));
+        RemainingLine = FEXCore::StringUtils::Trim(RemainingLine.substr(CurrentPos));
         CurrentPos = 0;
         if (RemainingLine.empty()) {
           // How did we get here?
@@ -495,7 +476,7 @@ class IRParser: public FEXCore::IR::IREmitter {
         else {
           while (!RemainingLine.empty()) {
             const size_t ArgEnd = RemainingLine.find(',');
-            std::string Arg = trim(RemainingLine.substr(0, ArgEnd));
+            std::string Arg = FEXCore::StringUtils::Trim(RemainingLine.substr(0, ArgEnd));
 
             Def.Args.emplace_back(std::move(Arg));
 
@@ -680,8 +661,8 @@ class IRParser: public FEXCore::IR::IREmitter {
 
 } // anon namespace
 
-std::unique_ptr<IREmitter> Parse(std::istream *in) {
-    auto parser = std::make_unique<IRParser>(in);
+std::unique_ptr<IREmitter> Parse(FEXCore::Utils::IntrusivePooledAllocator &ThreadAllocator, std::istream *in) {
+    auto parser = std::make_unique<IRParser>(ThreadAllocator, in);
 
     if (parser->Loaded) {
       return parser;

External/FEXCore/Source/Interface/IR/PassManager.cpp

@@ -6,6 +6,7 @@ desc: Defines which passes are run, and runs them
 $end_info$
 */
 
+#include "Interface/Context/Context.h"
 #include "Interface/IR/PassManager.h"
 #include "Interface/IR/Passes.h"
 #include "Interface/IR/Passes/RegisterAllocationPass.h"
@@ -15,7 +16,7 @@ $end_info$
 namespace FEXCore::IR {
 class IREmitter;
 
-void PassManager::AddDefaultPasses(bool InlineConstants, bool StaticRegisterAllocation) {
+void PassManager::AddDefaultPasses(FEXCore::Context::Context *ctx, bool InlineConstants, bool StaticRegisterAllocation) {
   FEX_CONFIG_OPT(DisablePasses, O0);
 
   if (!DisablePasses()) {
@@ -29,7 +30,7 @@ void PassManager::AddDefaultPasses(bool InlineConstants, bool StaticRegisterAllo
 
     InsertPass(CreateDeadStoreElimination());
     InsertPass(CreatePassDeadCodeElimination());
-    InsertPass(CreateConstProp(InlineConstants));
+    InsertPass(CreateConstProp(InlineConstants, ctx->HostFeatures.SupportsTSOImm9));
 
     ////// InsertPass(CreateDeadFlagCalculationEliminination());
 
@@ -48,7 +49,7 @@ void PassManager::AddDefaultPasses(bool InlineConstants, bool StaticRegisterAllo
 
   // If the IR is compacted post-RA then the node indexing gets messed up and the backend isn't able to find the register assigned to a node
   // Compact before IR, don't worry about RA generating spills/fills
-  InsertPass(CreateIRCompaction(), "Compaction");
+  InsertPass(CreateIRCompaction(ctx->OpDispatcherAllocator), "Compaction");
 }
 
 void PassManager::AddDefaultValidationPasses() {

External/FEXCore/Source/Interface/IR/PassManager.h

@@ -7,6 +7,7 @@ $end_info$
 #pragma once
 
 #include <FEXCore/Config/Config.h>
+#include <FEXCore/Utils/ThreadPoolAllocator.h>
 
 #include <functional>
 #include <memory>
@@ -39,7 +40,7 @@ protected:
 class PassManager final {
   friend class SyscallOptimization;
 public:
-  void AddDefaultPasses(bool InlineConstants, bool StaticRegisterAllocation);
+  void AddDefaultPasses(FEXCore::Context::Context *ctx, bool InlineConstants, bool StaticRegisterAllocation);
   void AddDefaultValidationPasses();
   Pass* InsertPass(std::unique_ptr<Pass> Pass, std::string Name = "") {
     Pass->RegisterPassManager(this);

External/FEXCore/Source/Interface/IR/Passes.h

@@ -2,18 +2,22 @@
 
 #include <memory>
 
+namespace FEXCore::Utils {
+class IntrusivePooledAllocator;
+}
+
 namespace FEXCore::IR {
 class Pass;
 class RegisterAllocationPass;
 class RegisterAllocationData;
 
-std::unique_ptr<FEXCore::IR::Pass> CreateConstProp(bool InlineConstants);
+std::unique_ptr<FEXCore::IR::Pass> CreateConstProp(bool InlineConstants, bool SupportsTSOImm9);
 std::unique_ptr<FEXCore::IR::Pass> CreateContextLoadStoreElimination();
 std::unique_ptr<FEXCore::IR::Pass> CreateSyscallOptimization();
 std::unique_ptr<FEXCore::IR::Pass> CreateDeadFlagCalculationEliminination();
 std::unique_ptr<FEXCore::IR::Pass> CreateDeadStoreElimination();
 std::unique_ptr<FEXCore::IR::Pass> CreatePassDeadCodeElimination();
-std::unique_ptr<FEXCore::IR::Pass> CreateIRCompaction();
+std::unique_ptr<FEXCore::IR::Pass> CreateIRCompaction(FEXCore::Utils::IntrusivePooledAllocator &Allocator);
 std::unique_ptr<FEXCore::IR::RegisterAllocationPass> CreateRegisterAllocationPass(FEXCore::IR::Pass* CompactionPass, bool OptimizeSRA);
 std::unique_ptr<FEXCore::IR::Pass> CreateStaticRegisterAllocationPass();
 std::unique_ptr<FEXCore::IR::Pass> CreateLongDivideEliminationPass();