Docs: Update for release FEX-2204

JITArm64: Get long divide out of the hot path

.github/workflows/ccpp.yml

@@ -31,7 +31,9 @@ jobs:
 
     - name : submodule checkout
       # Need to update submodules
-      run: git submodule update --init --depth 1
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
 
     - name: Clean Build Environment
       run: rm -Rf ${{runner.workspace}}/build
@@ -49,7 +51,7 @@ 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
+      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
 
     - name: Build
       working-directory: ${{runner.workspace}}/build

.gitmodules

@@ -1,7 +1,7 @@
 [submodule "External/vixl"]
 	shallow = true
 	path = External/vixl
-	url = https://github.com/Sonicadvance1/vixl.git
+	url = https://github.com/FEX-Emu/vixl.git
 [submodule "External/cpp-optparse"]
 	path = External/cpp-optparse
 	url = https://github.com/Sonicadvance1/cpp-optparse
@@ -45,3 +45,7 @@
 [submodule "External/Catch2"]
 	path = External/Catch2
 	url = https://github.com/catchorg/Catch2.git
+[submodule "External/robin-map"]
+	shallow = true
+	path = External/robin-map
+	url = https://github.com/Tessil/robin-map.git

CMakeLists.txt

@@ -19,6 +19,9 @@ option(ENABLE_JEMALLOC "Enables jemalloc allocator" TRUE)
 option(ENABLE_OFFLINE_TELEMETRY "Enables FEX offline telemetry" TRUE)
 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")
@@ -26,6 +29,7 @@ set (DATA_DIRECTORY "${CMAKE_INSTALL_PREFIX}/share/fex-emu" CACHE PATH "global d
 
 # These options are meant for package management
 set (TUNE_CPU "native" CACHE STRING "Override the CPU the build is tuned for")
+set (TUNE_ARCH "generic" CACHE STRING "Override the Arch the build is tuned for")
 set (OVERRIDE_VERSION "detect" CACHE STRING "Override the FEX version in the format of <MMYY>{.<REV>}")
 
 string(TOUPPER "${CMAKE_BUILD_TYPE}" CMAKE_BUILD_TYPE)
@@ -38,6 +42,11 @@ if (ENABLE_ASSERTIONS)
   add_definitions(-DASSERTIONS_ENABLED=1)
 endif()
 
+if (ENABLE_INTERPRETER)
+  message(STATUS "Interpreter enabled")
+  add_definitions(-DINTERPRETER_ENABLED=1)
+endif()
+
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/Bin)
@@ -72,10 +81,12 @@ if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
   add_definitions(-D_M_ARM_64=1)
 endif()
 
-find_program(CCACHE_PROGRAM ccache)
-if(CCACHE_PROGRAM)
-  message(STATUS "CCache enabled")
-  set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CCACHE_PROGRAM}")
+if (ENABLE_CCACHE)
+  find_program(CCACHE_PROGRAM ccache)
+  if(CCACHE_PROGRAM)
+    message(STATUS "CCache enabled")
+    set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CCACHE_PROGRAM}")
+  endif()
 endif()
 
 if (ENABLE_XRAY)
@@ -105,6 +116,13 @@ if (NOT ENABLE_OFFLINE_TELEMETRY)
   add_definitions(-DFEX_DISABLE_TELEMETRY=1)
 endif()
 
+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
+  set(ENABLE_JEMALLOC FALSE)
+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.")
@@ -258,6 +276,8 @@ set (CMAKE_LINKER_FLAGS_RELWITHDEBINFO "${CMAKE_LINKER_FLAGS_RELWITHDEBINFO} -fn
 set (CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fomit-frame-pointer")
 set (CMAKE_LINKER_FLAGS_RELEASE "${CMAKE_LINKER_FLAGS_RELEASE} -fomit-frame-pointer")
 
+include_directories(External/robin-map/include/)
+
 add_subdirectory(External/vixl/)
 include_directories(External/vixl/src/)
 
@@ -268,13 +288,9 @@ endif()
 
 find_package(PkgConfig REQUIRED)
 find_package(Python 3.0 REQUIRED COMPONENTS Interpreter)
-pkg_check_modules(XXHASH libxxhash>=0.8.0 QUIET)
 
-if (NOT XXHASH_FOUND)
-  message(STATUS "xxHash not found. Using Externals")
-  add_subdirectory(External/xxhash/)
-  include_directories(External/xxhash/)
-endif()
+add_subdirectory(External/xxhash/)
+include_directories(External/xxhash/)
 
 add_definitions(-Wno-trigraphs)
 add_definitions(-DGLOBAL_DATA_DIRECTORY="${DATA_DIRECTORY}/")
@@ -335,6 +351,15 @@ if(ENABLE_WERROR OR ENABLE_STRICT_WERROR)
   endif()
 endif()
 
+if (NOT TUNE_ARCH STREQUAL "generic")
+  check_cxx_compiler_flag("-march=${TUNE_ARCH}" COMPILER_SUPPORTS_ARCH_TYPE)
+  if(COMPILER_SUPPORTS_ARCH_TYPE)
+    add_compile_options("-march=${TUNE_ARCH}")
+  else()
+    message(FATAL_ERROR "Trying to compile arch type '${TUNE_ARCH}' but the compiler doesn't support this")
+  endif()
+endif()
+
 if (TUNE_CPU STREQUAL "native")
   if(_M_ARM_64)
     if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 999999.0)
@@ -468,30 +493,14 @@ endif()
 if (BUILD_THUNKS)
   add_subdirectory(ThunkLibs/Generator)
 
+  # Thunk targets for both host libraries and IDE integration
+  add_subdirectory(ThunkLibs/HostLibs)
+
+  # Thunk targets for IDE integration of guest code, only
+  add_subdirectory(ThunkLibs/GuestLibs)
+
+  # Thunk targets for guest libraries
   include(ExternalProject)
-
-  ExternalProject_Add(host-libs
-    PREFIX host-libs
-    SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/ThunkLibs/HostLibs"
-    BINARY_DIR "Host"
-    CMAKE_ARGS
-      "-DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}"
-      "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
-      "-DGENERATOR_EXE=$<TARGET_FILE:thunkgen>"
-    INSTALL_COMMAND ""
-    BUILD_ALWAYS ON
-    DEPENDS thunkgen
-  )
-
-  install(
-    CODE "MESSAGE(\"-- Installing: host-libs\")"
-    CODE "
-    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target ThunkHostsInstall
-    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Host
-    )"
-    DEPENDS host-libs
-  )
-
   ExternalProject_Add(guest-libs
     PREFIX guest-libs
     SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/ThunkLibs/GuestLibs"
@@ -511,7 +520,7 @@ if (BUILD_THUNKS)
   install(
     CODE "MESSAGE(\"-- Installing: guest-libs\")"
     CODE "
-    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target ThunkGuestsInstall
+    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target install
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest
     )"
     DEPENDS guest-libs

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/Scripts/json_ir_doc_generator.py

@@ -7,17 +7,12 @@ OpClasses = collections.OrderedDict()
 def get_ir_classes(ops, defines):
     global OpClasses
 
-    for op_key, op_vals in ops.items():
-        if not ("Last" in op_vals):
-            OpClass = "#Unknown"
+    for op_class, opslist in ops.items():
+        if not (op_class in OpClasses):
+            OpClasses[op_class] = []
 
-            if ("OpClass" in op_vals):
-                OpClass = op_vals["OpClass"]
-
-            if not (OpClass in OpClasses):
-                OpClasses[OpClass] = []
-
-            OpClasses[OpClass].append([op_key, op_vals])
+        for op, op_val in opslist.items():
+            OpClasses[op_class].append([op, op_val])
 
     # Sort the dictionary after we are done parsing it
     OpClasses = collections.OrderedDict(sorted(OpClasses.items()))
@@ -38,41 +33,9 @@ def print_ir_ops():
             op_key = op[0]
             op_vals = op[1]
             output_file.write("## %s\n" % (op_key))
-            HasDest = ("HasDest" in op_vals and op_vals["HasDest"] == True)
-            HasSSAArgs = ("SSAArgs" in op_vals and len(op_vals["SSAArgs"]) > 0)
-            HasSSAArgNames = "SSANames" in op_vals
-            HasArgs = "Args" in op_vals
-            SSAArgsCount = 0
-            ArgCount = 0
-            if (HasSSAArgs):
-                SSAArgsCount = int(op_vals["SSAArgs"])
-            if (HasArgs):
-                ArgCount = len(op_vals["Args"])
-
-            TotalArgsCount = SSAArgsCount + (ArgCount / 2)
 
             output_file.write(">")
-            if (HasDest):
-                output_file.write("%dest = ")
-
-            output_file.write("%s " % op_key)
-
-            ArgComma = (", ", "")
-            if (HasSSAArgs):
-                for i in range(0, SSAArgsCount):
-                    FinalArg = (i + 1) == TotalArgsCount
-                    if (HasSSAArgNames):
-                        output_file.write("%%%s%s" % (op_vals["SSANames"][i], ArgComma[FinalArg]))
-                    else:
-                        output_file.write("%%ssa%d%s" % (i, ArgComma[FinalArg]))
-
-            if (HasArgs):
-                Args = op_vals["Args"]
-                for i in range(0, ArgCount, 2):
-                    FinalArg = ((i / 2) + SSAArgsCount + 1) == TotalArgsCount
-                    data_type = Args[i]
-                    data_name = Args[i + 1]
-                    output_file.write("\<%s %s\>%s" % (data_type, data_name, ArgComma[FinalArg]))
+            output_file.write(op_key)
 
             output_file.write("\n\n")
 

External/FEXCore/Source/CMakeLists.txt

@@ -100,19 +100,7 @@ set (SRCS
   Interface/Core/Dispatcher/Dispatcher.cpp
   Interface/Core/Dispatcher/X86Dispatcher.cpp
   Interface/Core/Dispatcher/Arm64Dispatcher.cpp
-  Interface/Core/Interpreter/InterpreterCore.cpp
-  Interface/Core/Interpreter/InterpreterOps.cpp
-  Interface/Core/Interpreter/ALUOps.cpp
-  Interface/Core/Interpreter/AtomicOps.cpp
-  Interface/Core/Interpreter/BranchOps.cpp
-  Interface/Core/Interpreter/ConversionOps.cpp
-  Interface/Core/Interpreter/EncryptionOps.cpp
-  Interface/Core/Interpreter/F80Ops.cpp
-  Interface/Core/Interpreter/FlagOps.cpp
-  Interface/Core/Interpreter/MemoryOps.cpp
-  Interface/Core/Interpreter/MiscOps.cpp
-  Interface/Core/Interpreter/MoveOps.cpp
-  Interface/Core/Interpreter/VectorOps.cpp
+  Interface/Core/Interpreter/InterpreterFallbacks.cpp
   Interface/Core/X86Tables/BaseTables.cpp
   Interface/Core/X86Tables/DDDTables.cpp
   Interface/Core/X86Tables/EVEXTables.cpp
@@ -152,6 +140,23 @@ set (SRCS
   Utils/Threads.cpp
   )
 
+if (ENABLE_INTERPRETER)
+  list(APPEND SRCS
+    Interface/Core/Interpreter/InterpreterCore.cpp
+    Interface/Core/Interpreter/InterpreterOps.cpp
+    Interface/Core/Interpreter/ALUOps.cpp
+    Interface/Core/Interpreter/AtomicOps.cpp
+    Interface/Core/Interpreter/BranchOps.cpp
+    Interface/Core/Interpreter/ConversionOps.cpp
+    Interface/Core/Interpreter/EncryptionOps.cpp
+    Interface/Core/Interpreter/F80Ops.cpp
+    Interface/Core/Interpreter/FlagOps.cpp
+    Interface/Core/Interpreter/MemoryOps.cpp
+    Interface/Core/Interpreter/MiscOps.cpp
+    Interface/Core/Interpreter/MoveOps.cpp
+    Interface/Core/Interpreter/VectorOps.cpp)
+endif()
+
 if(_M_ARM_64)
   list(APPEND SRCS
     Interface/Core/ArchHelpers/Arm64.cpp)

External/FEXCore/Source/Common/JitSymbols.cpp

@@ -23,7 +23,7 @@ namespace FEXCore {
 
     // Linux perf format is very straightforward
     // `<HostPtr> <Size> <Name>\n`
-    fmt::print(fp.get(), "{:x} {:x} JIT_0x{:x}_{:x}\n", HostAddr, CodeSize, GuestAddr, HostAddr);
+    fmt::print(fp.get(), "{} {:x} JIT_0x{:x}_{}\n", HostAddr, CodeSize, GuestAddr, HostAddr);
   }
 
   void JITSymbols::Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name) {
@@ -31,7 +31,7 @@ namespace FEXCore {
 
     // Linux perf format is very straightforward
     // `<HostPtr> <Size> <Name>\n`
-    fmt::print(fp.get(), "{:x} {:x} {}_{:x}\n", HostAddr, CodeSize, Name, HostAddr);
+    fmt::print(fp.get(), "{} {:x} {}_{}\n", HostAddr, CodeSize, Name, HostAddr);
   }
 
   void JITSymbols::RegisterNamedRegion(const void *HostAddr, uint32_t CodeSize, std::string_view Name) {
@@ -39,7 +39,7 @@ namespace FEXCore {
 
     // Linux perf format is very straightforward
     // `<HostPtr> <Size> <Name>\n`
-    fmt::print(fp.get(), "{:x} {:x} {}\n", HostAddr, CodeSize, Name);
+    fmt::print(fp.get(), "{} {:x} {}\n", HostAddr, CodeSize, Name);
   }
 
   void JITSymbols::RegisterJITSpace(const void *HostAddr, uint32_t CodeSize) {
@@ -47,7 +47,7 @@ namespace FEXCore {
 
     // Linux perf format is very straightforward
     // `<HostPtr> <Size> <Name>\n`
-    fmt::print(fp.get(), "{:x} {:x} FEXJIT\n", HostAddr, CodeSize);
+    fmt::print(fp.get(), "{} {:x} FEXJIT\n", HostAddr, CodeSize);
   }
 
 } // namespace FEXCore

External/FEXCore/Source/Common/SoftFloat.h

@@ -57,51 +57,183 @@ struct X80SoftFloat {
 
   // Ops
   static X80SoftFloat FADD(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    faddp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_add(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FSUB(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fsubp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_sub(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FMUL(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fmulp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_mul(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FDIV(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fdivp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_div(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FREM(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
-    X80SoftFloat Rem = extF80_rem(lhs, rhs);
-    if (SignBit(Rem)) {
-      Rem = extF80_add(Rem, rhs);
-    }
-    else {
-      Rem.Sign = SignBit(lhs);
-    }
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fprem;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
 
-    return Rem;
+    return Result;
+#else
+    return extF80_rem(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FREM1(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fprem1;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_rem(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FRNDINT(X80SoftFloat const &lhs) {
     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;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fxtract;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     X80SoftFloat Tmp = lhs;
     Tmp.Exponent = 0x3FFF;
     Tmp.Sign = lhs.Sign;
     return Tmp;
+#endif
   }
 
   static X80SoftFloat FXTRACT_EXP(X80SoftFloat const &lhs) {
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fxtract;
+    ffreep %%st(0);
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     int32_t TrueExp = lhs.Exponent - ExponentBias;
     return i32_to_extF80(TrueExp);
+#endif
   }
 
   static void FCMP(X80SoftFloat const &lhs, X80SoftFloat const &rhs, bool *eq, bool *lt, bool *nan) {
@@ -112,61 +244,189 @@ struct X80SoftFloat {
 
   static X80SoftFloat FSCALE(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
     WARN_ONCE_FMT("x87: Application used FSCALE which may have accuracy problems");
-    X80SoftFloat Int = FRNDINT(rhs);
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fscale; # st0 = st0 * 2^(rdint(st1))
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
+    X80SoftFloat Int = FRNDINT(rhs, softfloat_round_minMag);
     BIGFLOAT Src2_d = Int;
     Src2_d = exp2l(Src2_d);
     X80SoftFloat Src2_X80 = Src2_d;
     X80SoftFloat Result = extF80_mul(lhs, Src2_X80);
     return Result;
+#endif
   }
 
   static X80SoftFloat F2XM1(X80SoftFloat const &lhs) {
     WARN_ONCE_FMT("x87: Application used F2XM1 which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    f2xm1; # st0 = 2^st(0) - 1
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     BIGFLOAT Src1_d = lhs;
     BIGFLOAT Result = exp2l(Src1_d);
     Result -= 1.0;
     return Result;
+#endif
   }
 
   static X80SoftFloat FYL2X(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
     WARN_ONCE_FMT("x87: Application used FYL2X which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st(1)
+    fldt %[lhs]; # st(0)
+    fyl2x; # st(1) * log2l(st(0))
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     BIGFLOAT Src1_d = lhs;
     BIGFLOAT Src2_d = rhs;
     BIGFLOAT Tmp = Src2_d * log2l(Src1_d);
     return Tmp;
+#endif
   }
 
   static X80SoftFloat FATAN(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
     WARN_ONCE_FMT("x87: Application used FATAN which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs];
+    fldt %[rhs];
+    fpatan;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     BIGFLOAT Src1_d = lhs;
     BIGFLOAT Src2_d = rhs;
     BIGFLOAT Tmp = atan2l(Src1_d, Src2_d);
     return Tmp;
+#endif
   }
 
   static X80SoftFloat FTAN(X80SoftFloat const &lhs) {
     WARN_ONCE_FMT("x87: Application used FTAN which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fptan;
+    ffreep %%st(0);
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     BIGFLOAT Src_d = lhs;
     Src_d = tanl(Src_d);
     return Src_d;
+#endif
   }
 
   static X80SoftFloat FSIN(X80SoftFloat const &lhs) {
     WARN_ONCE_FMT("x87: Application used FSIN which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fsin;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     BIGFLOAT Src_d = lhs;
     Src_d = sinl(Src_d);
     return Src_d;
+#endif
   }
 
   static X80SoftFloat FCOS(X80SoftFloat const &lhs) {
     WARN_ONCE_FMT("x87: Application used FCOS which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fcos;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     BIGFLOAT Src_d = lhs;
     Src_d = cosl(Src_d);
     return Src_d;
+#endif
   }
 
   static X80SoftFloat FSQRT(X80SoftFloat const &lhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fsqrt;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     return extF80_sqrt(lhs);
+#endif
   }
 
   operator float() const {

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 +371,6 @@ 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);
-    }
-
-    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 FindContainerPrefix() {
     // We only support pressure-vessel at the moment
     const static std::string ContainerManager = "/run/host/container-manager";
@@ -401,7 +379,7 @@ 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
@@ -434,7 +412,12 @@ namespace JSON {
 #else
       constexpr uint32_t MaxCoreNumber = 1;
 #endif
-      if (Core > MaxCoreNumber) {
+#ifdef INTERPRETER_ENABLED
+      constexpr uint32_t MinCoreNumber = 0;
+#else
+      constexpr uint32_t MinCoreNumber = 1;
+#endif
+      if (Core > MaxCoreNumber || Core < MinCoreNumber) {
         // Sanitize the core option by setting the core to the JIT if invalid
         FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_CORE, std::to_string(FEXCore::Config::CONFIG_IRJIT));
       }

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

@@ -535,7 +535,6 @@ uint64_t HandleCASPAL_ARMv8(void *_ucontext, void *_info, uint32_t Instr) {
 }
 
 bool HandleAtomicVectorStore(void *_ucontext, void *_info, uint32_t Instr) {
-  mcontext_t* mcontext = &reinterpret_cast<ucontext_t*>(_ucontext)->uc_mcontext;
   siginfo_t* info = reinterpret_cast<siginfo_t*>(_info);
 
   if (info->si_code != BUS_ADRALN) {

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

@@ -28,20 +28,32 @@ Arm64Emitter::Arm64Emitter(FEXCore::Context::Context *ctx, size_t size) : vixl::
   SetCPUFeatures(Features);
 }
 
-void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant) {
+void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant, bool NOPPad) {
   bool Is64Bit = Reg.IsX();
   int Segments = Is64Bit ? 4 : 2;
 
   if (Is64Bit && ((~Constant)>> 16) == 0) {
     movn(Reg, (~Constant) & 0xFFFF);
+
+    if (NOPPad) {
+      nop(); nop(); nop();
+    }
     return;
   }
 
+  int NumMoves = 1;
   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;
+    }
+  }
+
+  if (NOPPad) {
+    for (int i = NumMoves; i < Segments; ++i) {
+      nop();
     }
   }
 }
@@ -50,7 +62,7 @@ void Arm64Emitter::PushCalleeSavedRegisters() {
   // We need to save pairs of registers
   // We save r19-r30
   MemOperand PairOffset(sp, -16, PreIndex);
-  const std::array<std::pair<vixl::aarch64::XRegister, vixl::aarch64::XRegister>, 6> CalleeSaved = {{
+  const std::array<std::pair<vixl::aarch64::Register, vixl::aarch64::Register>, 6> CalleeSaved = {{
     {x19, x20},
     {x21, x22},
     {x23, x24},
@@ -113,7 +125,7 @@ void Arm64Emitter::PopCalleeSavedRegisters() {
   }
 
   MemOperand PairOffset(sp, 16, PostIndex);
-  const std::array<std::pair<vixl::aarch64::XRegister, vixl::aarch64::XRegister>, 6> CalleeSaved = {{
+  const std::array<std::pair<vixl::aarch64::Register, vixl::aarch64::Register>, 6> CalleeSaved = {{
     {x29, x30},
     {x27, x28},
     {x25, x26},
@@ -128,51 +140,75 @@ void Arm64Emitter::PopCalleeSavedRegisters() {
 }
 
 
-void Arm64Emitter::SpillStaticRegs(bool FPRs, uint32_t SpillMask) {
+void Arm64Emitter::SpillStaticRegs(bool FPRs, uint32_t GPRSpillMask, uint32_t FPRSpillMask) {
   if (StaticRegisterAllocation()) {
     for (size_t i = 0; i < SRA64.size(); i+=2) {
       auto Reg1 = SRA64[i];
       auto Reg2 = SRA64[i+1];
-      if (((1U << Reg1.GetCode()) & SpillMask) &&
-          ((1U << Reg2.GetCode()) & SpillMask)) {
+      if (((1U << Reg1.GetCode()) & GPRSpillMask) &&
+          ((1U << Reg2.GetCode()) & GPRSpillMask)) {
         stp(Reg1, Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
       }
-      else if (((1U << Reg1.GetCode()) & SpillMask)) {
+      else if (((1U << Reg1.GetCode()) & GPRSpillMask)) {
         str(Reg1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
       }
-      else if (((1U << Reg2.GetCode()) & SpillMask)) {
+      else if (((1U << Reg2.GetCode()) & GPRSpillMask)) {
         str(Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1])));
       }
     }
 
     if (FPRs) {
       for (size_t i = 0; i < SRAFPR.size(); i+=2) {
-        stp(SRAFPR[i].Q(), SRAFPR[i+1].Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+        auto Reg1 = SRAFPR[i];
+        auto Reg2 = SRAFPR[i+1];
+
+        if (((1U << Reg1.GetCode()) & FPRSpillMask) &&
+            ((1U << Reg2.GetCode()) & FPRSpillMask)) {
+          stp(Reg1.Q(), Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+        }
+        else if (((1U << Reg1.GetCode()) & FPRSpillMask)) {
+          str(Reg1.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+        }
+        else if (((1U << Reg2.GetCode()) & FPRSpillMask)) {
+          str(Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i+1][0])));
+        }
       }
     }
   }
 }
 
-void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t FillMask) {
+void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRFillMask) {
   if (StaticRegisterAllocation()) {
     for (size_t i = 0; i < SRA64.size(); i+=2) {
       auto Reg1 = SRA64[i];
       auto Reg2 = SRA64[i+1];
-      if (((1U << Reg1.GetCode()) & FillMask) &&
-          ((1U << Reg2.GetCode()) & FillMask)) {
+      if (((1U << Reg1.GetCode()) & GPRFillMask) &&
+          ((1U << Reg2.GetCode()) & GPRFillMask)) {
         ldp(Reg1, Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
       }
-      else if (((1U << Reg1.GetCode()) & FillMask)) {
+      else if (((1U << Reg1.GetCode()) & GPRFillMask)) {
         ldr(Reg1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
       }
-      else if (((1U << Reg2.GetCode()) & FillMask)) {
+      else if (((1U << Reg2.GetCode()) & GPRFillMask)) {
         ldr(Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1])));
       }
     }
 
     if (FPRs) {
       for (size_t i = 0; i < SRAFPR.size(); i+=2) {
-        ldp(SRAFPR[i].Q(), SRAFPR[i+1].Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+        auto Reg1 = SRAFPR[i];
+        auto Reg2 = SRAFPR[i+1];
+
+        if (((1U << Reg1.GetCode()) & FPRFillMask) &&
+            ((1U << Reg2.GetCode()) & FPRFillMask)) {
+          ldp(Reg1.Q(), Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+        }
+        else if (((1U << Reg1.GetCode()) & FPRFillMask)) {
+          ldr(Reg1.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+        }
+        else if (((1U << Reg2.GetCode()) & FPRFillMask)) {
+          ldr(Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i+1][0])));
+        }
       }
     }
   }

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

@@ -61,9 +61,15 @@ protected:
   Arm64Emitter(FEXCore::Context::Context *ctx, size_t size);
 
   vixl::aarch64::CPU CPU;
-  void LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant);
-  void SpillStaticRegs(bool FPRs = true, uint32_t SpillMask = ~0U);
-  void FillStaticRegs(bool FPRs = true, uint32_t FillMask = ~0U);
+  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);
+  void FillStaticRegs(bool FPRs = true, uint32_t GPRFillMask = ~0U, uint32_t FPRFillMask = ~0U);
+
+  static constexpr uint32_t CALLER_GPR_MASK = 0b0011'1111'1111'1111'1111;
+
+  // This isn't technically true because the lower 64-bits of v8..v15 are callee saved
+  // We can't guarantee only the lower 64bits are used so flush everything
+  static constexpr uint32_t CALLER_FPR_MASK = ~0U;
 
   void PushDynamicRegsAndLR();
   void PopDynamicRegsAndLR();

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

@@ -125,7 +125,8 @@ void CPUIDEmu::SetupHostHybridFlag() {
       // Only read 18 bytes for a 64bit value prefixed with 0x
       if (FEXCore::FileLoading::LoadFile(Data, MIDRPath, 18)) {
         uint64_t NewMIDR{};
-        if (FEXCore::StrConv::Conv(&Data.at(0), &NewMIDR)) {
+        std::string_view MIDRView(&Data.at(0), 18);
+        if (FEXCore::StrConv::Conv(MIDRView, &NewMIDR)) {
           if (MIDR != 0 && MIDR != NewMIDR) {
             // CPU mismatch, claim hybrid
             Hybrid = true;
@@ -403,6 +404,8 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_0h(uint32_t Leaf) {
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
   FEXCore::CPUID::FunctionResults Res{};
   uint32_t CoreCount = Cores();
+  // XXX: Enable once the rest of the SSE4.2 instructions are emulated
+  uint32_t SupportsSSE42 = CTX->HostFeatures.SupportsCRC && false ? 1 : 0;
 
   Res.eax = FAMILY_IDENTIFIER;
 
@@ -432,7 +435,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
     (0 << 17) | // Process-context identifiers
     (0 << 18) | // Prefetching from memory mapped device
     (1 << 19) | // SSE4.1
-    (0 << 20) | // SSE4.2
+    (SupportsSSE42 << 20) | // SSE4.2
     (0 << 21) | // X2APIC
     (1 << 22) | // MOVBE
     (1 << 23) | // POPCNT
@@ -442,7 +445,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
     (0 << 27) | // OSXSAVE
     (SUPPORTS_AVX << 28) | // AVX
     (0 << 29) | // F16C
-    (0 << 30) | // RDRAND
+    (CTX->HostFeatures.SupportsRAND << 30) | // RDRAND
     (0 << 31);  // Hypervisor always returns zero
 
   Res.edx =
@@ -644,7 +647,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
       (0 << 15) | // Intel Resource Directory Technology Allocation
       (0 << 16) | // Reserved
       (0 << 17) | // Reserved
-      (0 << 18) | // RDSEED
+      (CTX->HostFeatures.SupportsRAND << 18) | // RDSEED
       (1 << 19) | // ADCX and ADOX instructions
       (0 << 20) | // SMAP Supervisor mode access prevention and CLAC/STAC instructions
       (0 << 21) | // Reserved
@@ -809,6 +812,47 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_1Ah(uint32_t Leaf) {
   return Res;
 }
 
+// Hypervisor CPUID information leaf
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0000h(uint32_t Leaf) {
+  FEXCore::CPUID::FunctionResults Res{};
+  // Maximum supported hypervisor leafs
+  // We only expose the information leaf
+  //
+  // Common courtesy to follow VMWare's "Hypervisor CPUID Interface proposal"
+  // 4000_0000h - Information leaf. Advertising to the software which hypervisor this is
+  // 4000_0001h - 4000_000Fh - Hypervisor specific leafs. FEX can use these for anything
+  // 4000_0010h - 4000_00FFh - "Generic Leafs" - Try not to overwrite, other hypervisors might expect information in these
+  //
+  // 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 = 0x40000001;
+
+  // EBX, EDX, ECX become the hypervisor ID signature
+  constexpr static char HypervisorID[12] = "FEXIFEXIEMU";
+  memcpy(&Res.ebx, HypervisorID, sizeof(HypervisorID));
+  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{};
@@ -899,8 +943,8 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0001h(uint32_t Leaf) {
     (1 << 27) | // RDTSCP
     (0 << 28) | // Reserved
     (1 << 29) | // Long Mode
-    (0 << 30) | // 3DNow! Extensions
-    (0 << 31);  // 3DNow!
+    (1 << 30) | // 3DNow! Extensions
+    (1 << 31);  // 3DNow!
   return Res;
 }
 
@@ -1201,6 +1245,10 @@ void CPUIDEmu::Init(FEXCore::Context::Context *ctx) {
 #ifndef CPUID_AMD
   RegisterFunction(0x1A, &CPUIDEmu::Function_1Ah);
 #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);
   // Processor vendor

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

@@ -3,6 +3,7 @@
 #include <cstdint>
 #include <unordered_map>
 #include <utility>
+#include <vector>
 
 #include <FEXCore/Core/CPUID.h>
 #include <FEXCore/Config/Config.h>
@@ -78,6 +79,8 @@ private:
   FEXCore::CPUID::FunctionResults Function_0Dh(uint32_t Leaf);
   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/Core.cpp

@@ -190,6 +190,32 @@ namespace FEXCore::Context {
   }
 
   FEXCore::Core::InternalThreadState* Context::InitCore(FEXCore::CodeLoader *Loader) {
+    // Initialize the CPU core signal handlers
+    switch (Config.Core) {
+#ifdef INTERPRETER_ENABLED
+    case FEXCore::Config::CONFIG_INTERPRETER:
+      FEXCore::CPU::InitializeInterpreterSignalHandlers(this);
+      break;
+#endif
+    case FEXCore::Config::CONFIG_IRJIT:
+#if (_M_X86_64 && JIT_X86_64)
+      FEXCore::CPU::InitializeX86JITSignalHandlers(this);
+#elif (_M_ARM_64 && JIT_ARM64)
+      FEXCore::CPU::InitializeArm64JITSignalHandlers(this);
+#else
+      ERROR_AND_DIE_FMT("FEXCore has been compiled without a viable JIT core");
+#endif
+      break;
+    case FEXCore::Config::CONFIG_CUSTOM:
+      // Do nothing
+      break;
+    default:
+      ERROR_AND_DIE_FMT("Unknown core configuration");
+      break;
+    }
+
+    // Initialize GDBServer after the signal handlers are installed
+    // It may install its own handlers that need to be executed AFTER the CPU cores
     if (Config.GdbServer) {
       StartGdbServer();
     }
@@ -493,9 +519,11 @@ namespace FEXCore::Context {
 
     // Create CPU backend
     switch (Config.Core) {
+#ifdef INTERPRETER_ENABLED
     case FEXCore::Config::CONFIG_INTERPRETER:
       State->CPUBackend = FEXCore::CPU::CreateInterpreterCore(this, State, CompileThread);
       break;
+#endif
     case FEXCore::Config::CONFIG_IRJIT:
       State->PassManager->InsertRegisterAllocationPass(DoSRA);
 
@@ -869,10 +897,6 @@ namespace FEXCore::Context {
       StartAddr = _StartAddr;
       Length = _Length;
 
-      // Initialize metadata
-      DebugData->GuestCodeSize = TotalInstructionsLength;
-      DebugData->GuestInstructionCount = TotalInstructions;
-
       // Increment stats
       Thread->Stats.BlocksCompiled.fetch_add(1);
 
@@ -1120,10 +1144,16 @@ namespace FEXCore::Context {
 
   void Context::AddNamedRegion(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename) {
     IRCaptureCache.AddNamedRegion(Base, Size, Offset, filename);
+    if (DebugServer) {
+      DebugServer->AlertLibrariesChanged();
+    }
   }
 
   void Context::RemoveNamedRegion(uintptr_t Base, uintptr_t Size) {
     IRCaptureCache.RemoveNamedRegion(Base, Size);
+    if (DebugServer) {
+      DebugServer->AlertLibrariesChanged();
+    }
   }
 
   void ConfigureAOTGen(FEXCore::Core::InternalThreadState *Thread, std::set<uint64_t> *ExternalBranches, uint64_t SectionMaxAddress) {

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

@@ -53,10 +53,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   //    Ptr();
   // }
 
-  uint64_t VirtualMemorySize = Thread->LookupCache->GetVirtualMemorySize();
-  Literal l_VirtualMemory {VirtualMemorySize};
   Literal l_PagePtr {Thread->LookupCache->GetPagePointer()};
-  Literal l_L1Ptr {Thread->LookupCache->GetL1Pointer()};
   Literal l_CTX {reinterpret_cast<uintptr_t>(CTX)};
   Literal l_Sleep {reinterpret_cast<uint64_t>(SleepThread)};
   Literal l_CompileBlock {GetCompileBlockPtr()};
@@ -99,7 +96,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   auto RipReg = x2;
 
   // L1 Cache
-  ldr(x0, &l_L1Ptr);
+  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.L1Pointer)));
 
   and_(x3, RipReg, LookupCache::L1_ENTRIES_MASK);
   add(x0, x0, Operand(x3, Shift::LSL, 4));
@@ -121,11 +118,12 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   ldr(x0, &l_PagePtr);
 
   // Mask the address by the virtual address size so we can check for aliases
+  uint64_t VirtualMemorySize = Thread->LookupCache->GetVirtualMemorySize();
   if (std::popcount(VirtualMemorySize) == 1) {
-    and_(x3, RipReg, Thread->LookupCache->GetVirtualMemorySize() - 1);
+    and_(x3, RipReg, VirtualMemorySize - 1);
   }
   else {
-    ldr(x3, &l_VirtualMemory);
+    LoadConstant(x3, VirtualMemorySize);
     and_(x3, RipReg, x3);
   }
 
@@ -159,7 +157,7 @@ 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, &l_L1Ptr);
+      ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.L1Pointer)));
 
       and_(x1, RipReg, LookupCache::L1_ENTRIES_MASK);
       add(x0, x0, Operand(x1, Shift::LSL, 4));
@@ -438,9 +436,89 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     b(&LoopTop);
   }
 
-  place(&l_VirtualMemory);
+  // Long division helpers
+  uint64_t LUDIVHandler{};
+  uint64_t LDIVHandler{};
+  uint64_t LUREMHandler{};
+  uint64_t LREMHandler{};
+
+  {
+    LUDIVHandler = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR();
+
+    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::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();
+  }
+
+  {
+    LDIVHandler = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR();
+
+    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::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();
+  }
+
+  {
+    LUREMHandler = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR();
+
+    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::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();
+  }
+
+  {
+    LREMHandler = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR();
+
+    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::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_PagePtr);
-  place(&l_L1Ptr);
   place(&l_CTX);
   place(&l_Sleep);
   place(&l_CompileBlock);
@@ -461,6 +539,24 @@ 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();
+    Pointers.LUDIVHandler = LUDIVHandler;
+    Pointers.LDIVHandler = LDIVHandler;
+    Pointers.LUREMHandler = LUREMHandler;
+    Pointers.LREMHandler = LREMHandler;
+  }
 }
 
 void Arm64Dispatcher::SpillSRA(void *ucontext, uint32_t IgnoreMask) {

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

@@ -16,9 +16,9 @@
 
 #include <atomic>
 #include <condition_variable>
-#include <bits/types/siginfo_t.h>
 #include <csignal>
 #include <cstring>
+#include <signal.h>
 
 namespace FEXCore::CPU {
 

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

@@ -5,8 +5,8 @@
 #include "Interface/Context/Context.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
 
-#include <bits/types/stack_t.h>
 #include <cstdint>
+#include <signal.h>
 #include <stddef.h>
 #include <stack>
 #include <tuple>

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

@@ -95,7 +95,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     mov(rdx, qword [STATE + offsetof(FEXCore::Core::CPUState, rip)]);
 
     // L1 Cache
-    mov(r13, Thread->LookupCache->GetL1Pointer());
+    mov(r13, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
     mov(rax, rdx);
 
     and_(rax, LookupCache::L1_ENTRIES_MASK);
@@ -114,8 +114,9 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     mov(r13, Thread->LookupCache->GetPagePointer());
 
     // Full lookup
+    uint64_t VirtualMemorySize = Thread->LookupCache->GetVirtualMemorySize();
     mov(rax, rdx);
-    mov(rbx, Thread->LookupCache->GetVirtualMemorySize() - 1);
+    mov(rbx, VirtualMemorySize - 1);
     and_(rax, rbx);
     shr(rax, 12);
 
@@ -142,8 +143,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     je(NoBlock);
 
     // Update L1
-
-    mov(r13, Thread->LookupCache->GetL1Pointer());
+    mov(r13, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
     mov(rcx, rdx);
     and_(rcx, LookupCache::L1_ENTRIES_MASK);
     shl(rcx, 1);
@@ -252,8 +252,7 @@ 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
-    mov(rax, reinterpret_cast<uint64_t>(&SignalHandlerRefCounter));
-    add(dword [rax], 1);
+    add(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalHandlerRefCountPointer)], 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
@@ -337,6 +336,20 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
   if (CTX->Config.GlobalJITNaming()) {
     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();
+  }
 }
 
 X86Dispatcher::~X86Dispatcher() {

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

@@ -583,8 +583,11 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
         return false;
     }
     else {
-      auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
-      (this->*Disp)(&NonGPR, ModRM);
+      // Only decode if we haven't pre-decoded
+      if (NonGPR.IsNone()) {
+        auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
+        (this->*Disp)(&NonGPR, ModRM);
+      }
     }
 
     return true;
@@ -857,7 +860,7 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
     case 0x0F: {// Escape Op
       uint8_t EscapeOp = ReadByte();
       switch (EscapeOp) {
-      case 0x0F: { // 3DNow!
+      case 0x0F: [[unlikely]] { // 3DNow!
         // 3DNow! Instruction Encoding: 0F 0F [ModRM] [SIB] [Displacement] [Opcode]
         // Decode ModRM
         uint8_t ModRMByte = ReadByte();
@@ -870,8 +873,12 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
         const bool Has16BitAddressing = !CTX->Config.Is64BitMode &&
           DecodeInst->Flags & DecodeFlags::FLAG_ADDRESS_SIZE;
 
-        auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
-        (this->*Disp)(&DecodeInst->Src[0], ModRM);
+        // All 3DNow! instructions have the second argument as the rm handler
+        // We need to decode it upfront to get the displacement out of the way
+        if (ModRM.mod != 0b11) {
+          auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
+          (this->*Disp)(&DecodeInst->Src[0], ModRM);
+        }
 
         // Take a peek at the op just past the displacement
         uint8_t LocalOp = ReadByte();
@@ -880,20 +887,19 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
       }
       case 0x38: { // F38 Table!
         constexpr uint16_t PF_38_NONE = 0;
-        constexpr uint16_t PF_38_66 = 1;
-        constexpr uint16_t PF_38_F2 = 2;
-        constexpr uint16_t PF_38_F3 = 3;
+        constexpr uint16_t PF_38_66   = (1U << 0);
+        constexpr uint16_t PF_38_F2   = (1U << 1);
+        constexpr uint16_t PF_38_F3   = (1U << 2);
 
         uint16_t Prefix = PF_38_NONE;
-        if (DecodeInst->LastEscapePrefix == 0xF2) {
-          // Repeat prefix or instruction-specific
-          Prefix = PF_38_F2;
-        } else if (DecodeInst->LastEscapePrefix == 0xF3) {
-          // Repeat prefix or instruction-specific
-          Prefix = PF_38_F3;
-        } else if (DecodeInst->LastEscapePrefix == 0x66) {
-          // Operand size
-          Prefix = PF_38_66;
+        if (DecodeInst->Flags & DecodeFlags::FLAG_OPERAND_SIZE) {
+          Prefix |= PF_38_66;
+        }
+        if (DecodeInst->Flags & DecodeFlags::FLAG_REPNE_PREFIX) {
+          Prefix |= PF_38_F2;
+        }
+        if (DecodeInst->Flags & DecodeFlags::FLAG_REP_PREFIX) {
+          Prefix |= PF_38_F3;
         }
 
         uint16_t LocalOp = (Prefix << 8) | ReadByte();

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

@@ -35,7 +35,7 @@ public:
 
   uint64_t DecodedMinAddress {};
   uint64_t DecodedMaxAddress {~0ULL};
-  
+
   void SetSectionMaxAddress(uint64_t v) { SectionMaxAddress = v; }
   void SetExternalBranches(std::set<uint64_t> *v) { ExternalBranches = v; }
 private:

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

@@ -53,7 +53,10 @@ HostFeatures::HostFeatures() {
 #ifdef _M_ARM_64
   auto Features = vixl::CPUFeatures::InferFromOS();
   SupportsAES = Features.Has(vixl::CPUFeatures::Feature::kAES);
+  SupportsCRC = Features.Has(vixl::CPUFeatures::Feature::kCRC32);
   SupportsAtomics = Features.Has(vixl::CPUFeatures::Feature::kAtomics);
+  SupportsRAND = Features.Has(vixl::CPUFeatures::Feature::kRNG);
+
   // Only supported when FEAT_AFP is supported
   SupportsFlushInputsToZero = Features.Has(vixl::CPUFeatures::Feature::kAFP);
 
@@ -78,6 +81,19 @@ HostFeatures::HostFeatures() {
 #ifdef _M_X86_64
   Xbyak::util::Cpu Features{};
   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);
+
+  // 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;
 #else

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

@@ -15,9 +15,11 @@ class HostFeatures final {
     uint32_t DCacheLineSize{};
     uint32_t ICacheLineSize{};
     bool SupportsAES{};
+    bool SupportsCRC{};
     bool SupportsCLZERO{};
     bool SupportsAtomics{};
     bool SupportsRCPC{};
+    bool SupportsRAND{};
 
     // Float exception behaviour
     bool SupportsFlushInputsToZero{};

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

@@ -18,7 +18,7 @@ namespace FEXCore::CPU {
 DEF_OP(TruncElementPair) {
   auto Op = IROp->C<IR::IROp_TruncElementPair>();
 
-  switch (Op->Size) {
+  switch (IROp->Size) {
     case 4: {
       uint64_t *Src = GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
       uint64_t Result{};
@@ -27,7 +27,7 @@ DEF_OP(TruncElementPair) {
       GD = Result;
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unhandled Truncation size: {}", Op->Size); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled Truncation size: {}", IROp->Size); break;
   }
 }
 
@@ -38,7 +38,13 @@ DEF_OP(Constant) {
 
 DEF_OP(EntrypointOffset) {
   auto Op = IROp->C<IR::IROp_EntrypointOffset>();
-  GD = Data->CurrentEntry + Op->Offset;
+  uint64_t Mask = ~0ULL;
+  uint8_t OpSize = IROp->Size;
+  if (OpSize == 4) {
+    Mask = 0xFFFF'FFFFULL;
+  }
+
+  GD = (Data->CurrentEntry + Op->Offset) & Mask;
 }
 
 DEF_OP(InlineConstant) {
@@ -835,9 +841,8 @@ DEF_OP(Bfi) {
 
 DEF_OP(Bfe) {
   auto Op = IROp->C<IR::IROp_Bfe>();
-  uint8_t OpSize = IROp->Size;
 
-  LOGMAN_THROW_A_FMT(OpSize <= 8, "OpSize is too large for BFE: {}", OpSize);
+  LOGMAN_THROW_A_FMT(IROp->Size <= 8, "OpSize is too large for BFE: {}", IROp->Size);
   uint64_t SourceMask = (1ULL << Op->Width) - 1;
   if (Op->Width == 64)
     SourceMask = ~0ULL;
@@ -848,9 +853,8 @@ DEF_OP(Bfe) {
 
 DEF_OP(Sbfe) {
   auto Op = IROp->C<IR::IROp_Sbfe>();
-  uint8_t OpSize = IROp->Size;
 
-  LOGMAN_THROW_A_FMT(OpSize <= 8, "OpSize is too large for SBFE: {}", OpSize);
+  LOGMAN_THROW_A_FMT(IROp->Size <= 8, "OpSize is too large for SBFE: {}", IROp->Size);
   int64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
   uint64_t ShiftLeftAmount = (64 - (Op->Width + Op->lsb));
   uint64_t ShiftRightAmount = ShiftLeftAmount + Op->lsb;
@@ -889,15 +893,14 @@ DEF_OP(Select) {
 
 DEF_OP(VExtractToGPR) {
   auto Op = IROp->C<IR::IROp_VExtractToGPR>();
-  uint8_t OpSize = IROp->Size;
 
   uint32_t SourceSize = GetOpSize(Data->CurrentIR, Op->Header.Args[0]);
 
-  LOGMAN_THROW_A_FMT(OpSize <= 16, "OpSize is too large for VExtractToGPR: {}", OpSize);
+  LOGMAN_THROW_A_FMT(IROp->Size <= 16, "OpSize is too large for VExtractToGPR: {}", IROp->Size);
 
   if (SourceSize == 16) {
     __uint128_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-    uint64_t Shift = Op->Header.ElementSize * Op->Idx * 8;
+    uint64_t Shift = Op->Header.ElementSize * Op->Index * 8;
     if (Op->Header.ElementSize == 8)
       SourceMask = ~0ULL;
 
@@ -908,7 +911,7 @@ DEF_OP(VExtractToGPR) {
   }
   else {
     uint64_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-    uint64_t Shift = Op->Header.ElementSize * Op->Idx * 8;
+    uint64_t Shift = Op->Header.ElementSize * Op->Index * 8;
     if (Op->Header.ElementSize == 8)
       SourceMask = ~0ULL;
 

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

@@ -313,30 +313,29 @@ uint64_t AtomicCompareAndSwap(uint64_t expected, uint64_t desired, uint64_t *add
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(CASPair) {
   auto Op = IROp->C<IR::IROp_CASPair>();
-  uint8_t OpSize = IROp->Size;
 
   // Size is the size of each pair element
-  switch (OpSize) {
+  switch (IROp->ElementSize) {
     case 4: {
       GD = AtomicCompareAndSwap(
-        *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint64_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint64_t**>(Data->SSAData, Op->Addr)
       );
       break;
     }
     case 8: {
-      std::atomic<__uint128_t> *MemData = *GetSrc<std::atomic<__uint128_t> **>(Data->SSAData, Op->Header.Args[2]);
+      std::atomic<__uint128_t> *MemData = *GetSrc<std::atomic<__uint128_t> **>(Data->SSAData, Op->Addr);
 
-      __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-      __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Expected);
+      __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Desired);
 
       __uint128_t Expected = Src1;
       bool Result = MemData->compare_exchange_strong(Expected, Src2);
       memcpy(GDP, Result ? &Src1 : &Expected, 16);
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown CAS size: {}", OpSize); break;
+    default: LOGMAN_MSG_A_FMT("Unknown CAS size: {}", IROp->ElementSize); break;
   }
 }
 
@@ -347,33 +346,33 @@ DEF_OP(CAS) {
   switch (OpSize) {
     case 1: {
       GD = AtomicCompareAndSwap(
-        *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint8_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint8_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint8_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint8_t**>(Data->SSAData, Op->Addr)
       );
       break;
     }
     case 2: {
       GD = AtomicCompareAndSwap(
-        *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint16_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint16_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint16_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint16_t**>(Data->SSAData, Op->Addr)
       );
       break;
     }
     case 4: {
       GD = AtomicCompareAndSwap(
-        *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint32_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint32_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint32_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint32_t**>(Data->SSAData, Op->Addr)
       );
       break;
     }
     case 8: {
       GD = AtomicCompareAndSwap(
-        *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint64_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint64_t**>(Data->SSAData, Op->Addr)
       );
       break;
     }
@@ -385,26 +384,26 @@ DEF_OP(AtomicAdd) {
   auto Op = IROp->C<IR::IROp_AtomicAdd>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       *MemData += Src;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       *MemData += Src;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       *MemData += Src;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       *MemData += Src;
       break;
     }
@@ -416,26 +415,26 @@ DEF_OP(AtomicSub) {
   auto Op = IROp->C<IR::IROp_AtomicSub>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       *MemData -= Src;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       *MemData -= Src;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       *MemData -= Src;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       *MemData -= Src;
       break;
     }
@@ -447,26 +446,26 @@ DEF_OP(AtomicAnd) {
   auto Op = IROp->C<IR::IROp_AtomicAnd>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       *MemData &= Src;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       *MemData &= Src;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       *MemData &= Src;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       *MemData &= Src;
       break;
     }
@@ -478,26 +477,26 @@ DEF_OP(AtomicOr) {
   auto Op = IROp->C<IR::IROp_AtomicOr>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       *MemData |= Src;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       *MemData |= Src;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       *MemData |= Src;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       *MemData |= Src;
       break;
     }
@@ -509,26 +508,26 @@ DEF_OP(AtomicXor) {
   auto Op = IROp->C<IR::IROp_AtomicXor>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       *MemData ^= Src;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       *MemData ^= Src;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       *MemData ^= Src;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       *MemData ^= Src;
       break;
     }
@@ -540,29 +539,29 @@ DEF_OP(AtomicSwap) {
   auto Op = IROp->C<IR::IROp_AtomicSwap>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       uint8_t Previous = MemData->exchange(Src);
       GD = Previous;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       uint16_t Previous = MemData->exchange(Src);
       GD = Previous;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       uint32_t Previous = MemData->exchange(Src);
       GD = Previous;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       uint64_t Previous = MemData->exchange(Src);
       GD = Previous;
       break;
@@ -575,29 +574,29 @@ DEF_OP(AtomicFetchAdd) {
   auto Op = IROp->C<IR::IROp_AtomicFetchAdd>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       uint8_t Previous = MemData->fetch_add(Src);
       GD = Previous;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       uint16_t Previous = MemData->fetch_add(Src);
       GD = Previous;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       uint32_t Previous = MemData->fetch_add(Src);
       GD = Previous;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       uint64_t Previous = MemData->fetch_add(Src);
       GD = Previous;
       break;
@@ -610,29 +609,29 @@ DEF_OP(AtomicFetchSub) {
   auto Op = IROp->C<IR::IROp_AtomicFetchSub>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       uint8_t Previous = MemData->fetch_sub(Src);
       GD = Previous;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       uint16_t Previous = MemData->fetch_sub(Src);
       GD = Previous;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       uint32_t Previous = MemData->fetch_sub(Src);
       GD = Previous;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       uint64_t Previous = MemData->fetch_sub(Src);
       GD = Previous;
       break;
@@ -645,29 +644,29 @@ DEF_OP(AtomicFetchAnd) {
   auto Op = IROp->C<IR::IROp_AtomicFetchAnd>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       uint8_t Previous = MemData->fetch_and(Src);
       GD = Previous;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       uint16_t Previous = MemData->fetch_and(Src);
       GD = Previous;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       uint32_t Previous = MemData->fetch_and(Src);
       GD = Previous;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       uint64_t Previous = MemData->fetch_and(Src);
       GD = Previous;
       break;
@@ -680,29 +679,29 @@ DEF_OP(AtomicFetchOr) {
   auto Op = IROp->C<IR::IROp_AtomicFetchOr>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       uint8_t Previous = MemData->fetch_or(Src);
       GD = Previous;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       uint16_t Previous = MemData->fetch_or(Src);
       GD = Previous;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       uint32_t Previous = MemData->fetch_or(Src);
       GD = Previous;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       uint64_t Previous = MemData->fetch_or(Src);
       GD = Previous;
       break;
@@ -715,29 +714,29 @@ DEF_OP(AtomicFetchXor) {
   auto Op = IROp->C<IR::IROp_AtomicFetchXor>();
   switch (IROp->Size) {
     case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
       uint8_t Previous = MemData->fetch_xor(Src);
       GD = Previous;
       break;
     }
     case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
       uint16_t Previous = MemData->fetch_xor(Src);
       GD = Previous;
       break;
     }
     case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
       uint32_t Previous = MemData->fetch_xor(Src);
       GD = Previous;
       break;
     }
     case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
       uint64_t Previous = MemData->fetch_xor(Src);
       GD = Previous;
       break;
@@ -751,22 +750,22 @@ DEF_OP(AtomicFetchNeg) {
   switch (IROp->Size) {
     case 1: {
       using Type = uint8_t;
-      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Header.Args[0]));
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
       break;
     }
     case 2: {
       using Type = uint16_t;
-      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Header.Args[0]));
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
       break;
     }
     case 4: {
       using Type = uint32_t;
-      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Header.Args[0]));
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
       break;
     }
     case 8: {
       using Type = uint64_t;
-      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Header.Args[0]));
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
       break;
     }
     default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);

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

@@ -33,10 +33,6 @@ DEF_OP(GuestCallIndirect) {
   LogMan::Msg::DFmt("Unimplemented");
 }
 
-DEF_OP(GuestReturn) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
 DEF_OP(SignalReturn) {
   SignalReturn(Data->State);
 }

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

@@ -19,7 +19,7 @@ DEF_OP(VInsGPR) {
   __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
   __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
 
-  uint64_t Offset = Op->Index * Op->Header.ElementSize * 8;
+  uint64_t Offset = Op->DestIdx * Op->Header.ElementSize * 8;
   __uint128_t Mask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
   if (Op->Header.ElementSize == 8) {
     Mask = ~0ULL;

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

@@ -298,6 +298,63 @@ namespace AES {
   }
 }
 
+namespace CRC32 {
+  // CRC32 per byte lookup table.
+  constexpr std::array<uint32_t, 256> CRC32CTable = []() consteval {
+    std::array<uint32_t, 256> Table{};
+
+    // Clang 11.x doesn't support bitreverse as a consteval
+    // constexpr uint32_t Polynomial = 0x1EDC6F41;
+    constexpr uint32_t PolynomialRev = 0x82F63B78; //__builtin_bitreverse32(Polynomial);
+
+    for (size_t Char = 0; Char < std::size(Table); ++Char) {
+      uint32_t CurrentChar = Char;
+      for (size_t i = 0; i < 8; ++i) {
+        if (CurrentChar & 1) {
+          CurrentChar = (CurrentChar >> 1) ^ PolynomialRev;
+        }
+        else {
+          CurrentChar >>= 1;
+        }
+      }
+      Table[Char] = CurrentChar;
+    }
+
+    return Table;
+  }();
+
+  uint32_t crc32cb(uint32_t Accumulator, uint8_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ data] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32ch(uint32_t Accumulator, uint16_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 8) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32cw(uint32_t Accumulator, uint32_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  8) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 16) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 24) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32cx(uint32_t Accumulator, uint64_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  8) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 16) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 24) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 32) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 40) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 48) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 56) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+}
+
 namespace FEXCore::CPU {
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 
@@ -429,6 +486,33 @@ DEF_OP(AESKeyGenAssist) {
   memcpy(GDP, &Tmp, sizeof(Tmp));
 }
 
+DEF_OP(CRC32) {
+  auto Op = IROp->C<IR::IROp_CRC32>();
+  uint32_t Src1 = *GetSrc<uint32_t*>(Data->SSAData, Op->Src1);
+  uint8_t *Src2 = GetSrc<uint8_t*>(Data->SSAData, Op->Src2);
+  uint32_t Tmp{};
+
+  switch (Op->SrcSize) {
+    case 1:
+      Tmp = CRC32::crc32cb(Src1, *(uint8_t*)Src2);
+      break;
+    case 2:
+      Tmp = CRC32::crc32ch(Src1, *(uint16_t*)Src2);
+      break;
+    case 4:
+      Tmp = CRC32::crc32cw(Src1, *(uint32_t*)Src2);
+      break;
+    case 8:
+      Tmp = CRC32::crc32cx(Src1, *(uint64_t*)Src2);
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown CRC32C size: {}", Op->SrcSize);
+      break;
+
+  }
+  memcpy(GDP, &Tmp, sizeof(Tmp));
+}
+
 #undef DEF_OP
 
 } // namespace FEXCore::CPU

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

@@ -158,7 +158,7 @@ DEF_OP(F80CVTINT) {
 DEF_OP(F80CVTTO) {
   auto Op = IROp->C<IR::IROp_F80CVTTo>();
 
-  switch (Op->Size) {
+  switch (Op->SrcSize) {
     case 4: {
       float Src = *GetSrc<float *>(Data->SSAData, Op->Header.Args[0]);
       X80SoftFloat Tmp = Src;
@@ -171,14 +171,14 @@ DEF_OP(F80CVTTO) {
       memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
       break;
     }
-    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->Size);
+    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->SrcSize);
   }
 }
 
 DEF_OP(F80CVTTOINT) {
   auto Op = IROp->C<IR::IROp_F80CVTToInt>();
 
-  switch (Op->Size) {
+  switch (Op->SrcSize) {
     case 2: {
       int16_t Src = *GetSrc<int16_t*>(Data->SSAData, Op->Header.Args[0]);
       X80SoftFloat Tmp = Src;
@@ -191,7 +191,7 @@ DEF_OP(F80CVTTOINT) {
       memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
       break;
     }
-    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->Size);
+    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->SrcSize);
   }
 }
 
@@ -323,7 +323,7 @@ DEF_OP(F80BCDLOAD) {
 
 DEF_OP(F80BCDSTORE) {
   auto Op = IROp->C<IR::IROp_F80BCDStore>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
+  X80SoftFloat Src1 = X80SoftFloat::FRNDINT(*GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]));
   bool Negative = Src1.Sign;
 
   // Clear the Sign bit

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

@@ -227,6 +227,8 @@ struct OpHandlers<IR::OP_F80BCDSTORE> {
   static X80SoftFloat handle(X80SoftFloat Src1) {
     bool Negative = Src1.Sign;
 
+    Src1 = X80SoftFloat::FRNDINT(Src1);
+
     // Clear the Sign bit
     Src1.Sign = 0;
 

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

@@ -37,6 +37,10 @@ public:
 
   void CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread);
 
+  static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
+
+  bool NeedsRetainedIRCopy() const override { return true; }
+
 private:
   FEXCore::Context::Context *CTX;
   FEXCore::Core::InternalThreadState *State;

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

@@ -11,7 +11,6 @@
 #include <FEXCore/Utils/LogManager.h>
 
 #include <memory>
-#include <bits/types/stack_t.h>
 #include <signal.h>
 #include <stdint.h>
 #include <unordered_map>
@@ -42,25 +41,28 @@ InterpreterCore::InterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::
   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) {
+  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);
 
 #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);
-    };
+  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);
+  };
 
-    for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
-      CTX->SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
-    }
+  for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
+    CTX->SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
   }
 }
 
@@ -72,4 +74,8 @@ std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx
   return std::make_unique<InterpreterCore>(ctx, Thread, CompileThread);
 }
 
+void InitializeInterpreterSignalHandlers(FEXCore::Context::Context *CTX) {
+  InterpreterCore::InitializeSignalHandlers(CTX);
+}
+
 }

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

@@ -17,4 +17,6 @@ class CPUBackend;
                                                                 FEXCore::Core::InternalThreadState *Thread,
                                                                 bool CompileThread);
 
+void InitializeInterpreterSignalHandlers(FEXCore::Context::Context *CTX);
+
 } // namespace FEXCore::CPU

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

@@ -0,0 +1,272 @@
+#include "FEXCore/Core/CoreState.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/F80Ops.h"
+
+#include <cstddef>
+#include <cstdint>
+
+namespace FEXCore::CPU {
+
+template<typename R, typename... Args>
+static FallbackInfo GetFallbackInfo(R(*fn)(Args...), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_UNKNOWN, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(float), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F32, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F64, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int16_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_I16, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(void(*fn)(uint16_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_VOID_U16, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int32_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_I32, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(float(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F32_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int16_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I16_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int32_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I32_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int64_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I64_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(uint64_t(*fn)(X80SoftFloat, X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I64_F80_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat, X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F80_F80, (void*)fn, HandlerIndex};
+}
+
+void InterpreterOps::FillFallbackIndexPointers(uint64_t *Info) {
+  Info[Core::OPINDEX_F80LOADFCW] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle, Core::OPINDEX_F80LOADFCW).fn);
+  Info[Core::OPINDEX_F80CVTTO_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle4, Core::OPINDEX_F80CVTTO_4).fn);
+  Info[Core::OPINDEX_F80CVTTO_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle8, Core::OPINDEX_F80CVTTO_8).fn);
+  Info[Core::OPINDEX_F80CVT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle4, Core::OPINDEX_F80CVT_4).fn);
+  Info[Core::OPINDEX_F80CVT_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle8, Core::OPINDEX_F80CVT_8).fn);
+  Info[Core::OPINDEX_F80CVTINT_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2, Core::OPINDEX_F80CVTINT_2).fn);
+  Info[Core::OPINDEX_F80CVTINT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4, Core::OPINDEX_F80CVTINT_4).fn);
+  Info[Core::OPINDEX_F80CVTINT_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8, Core::OPINDEX_F80CVTINT_8).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t, Core::OPINDEX_F80CVTINT_TRUNC2).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t, Core::OPINDEX_F80CVTINT_TRUNC4).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t, Core::OPINDEX_F80CVTINT_TRUNC8).fn);
+  Info[Core::OPINDEX_F80CMP_0] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>, Core::OPINDEX_F80CMP_0).fn);
+  Info[Core::OPINDEX_F80CMP_1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>, Core::OPINDEX_F80CMP_1).fn);
+  Info[Core::OPINDEX_F80CMP_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<2>, Core::OPINDEX_F80CMP_2).fn);
+  Info[Core::OPINDEX_F80CMP_3] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<3>, Core::OPINDEX_F80CMP_3).fn);
+  Info[Core::OPINDEX_F80CMP_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<4>, Core::OPINDEX_F80CMP_4).fn);
+  Info[Core::OPINDEX_F80CMP_5] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<5>, Core::OPINDEX_F80CMP_5).fn);
+  Info[Core::OPINDEX_F80CMP_6] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<6>, Core::OPINDEX_F80CMP_6).fn);
+  Info[Core::OPINDEX_F80CMP_7] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<7>, Core::OPINDEX_F80CMP_7).fn);
+  Info[Core::OPINDEX_F80CVTTOINT_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle2, Core::OPINDEX_F80CVTTOINT_2).fn);
+  Info[Core::OPINDEX_F80CVTTOINT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle4, Core::OPINDEX_F80CVTTOINT_4).fn);
+
+  // Unary
+  Info[Core::OPINDEX_F80ROUND] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ROUND>::handle, Core::OPINDEX_F80ROUND).fn);
+  Info[Core::OPINDEX_F80F2XM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80F2XM1>::handle, Core::OPINDEX_F80F2XM1).fn);
+  Info[Core::OPINDEX_F80TAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80TAN>::handle, Core::OPINDEX_F80TAN).fn);
+  Info[Core::OPINDEX_F80SQRT] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SQRT>::handle, Core::OPINDEX_F80SQRT).fn);
+  Info[Core::OPINDEX_F80SIN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SIN>::handle, Core::OPINDEX_F80SIN).fn);
+  Info[Core::OPINDEX_F80COS] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80COS>::handle, Core::OPINDEX_F80COS).fn);
+  Info[Core::OPINDEX_F80XTRACT_EXP] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80XTRACT_EXP>::handle, Core::OPINDEX_F80XTRACT_EXP).fn);
+  Info[Core::OPINDEX_F80XTRACT_SIG] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80XTRACT_SIG>::handle, Core::OPINDEX_F80XTRACT_SIG).fn);
+  Info[Core::OPINDEX_F80BCDSTORE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80BCDSTORE>::handle, Core::OPINDEX_F80BCDSTORE).fn);
+  Info[Core::OPINDEX_F80BCDLOAD] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80BCDLOAD>::handle, Core::OPINDEX_F80BCDLOAD).fn);
+
+  // Binary
+  Info[Core::OPINDEX_F80ADD] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ADD>::handle, Core::OPINDEX_F80ADD).fn);
+  Info[Core::OPINDEX_F80SUB] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SUB>::handle, Core::OPINDEX_F80SUB).fn);
+  Info[Core::OPINDEX_F80MUL] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80MUL>::handle, Core::OPINDEX_F80MUL).fn);
+  Info[Core::OPINDEX_F80DIV] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80DIV>::handle, Core::OPINDEX_F80DIV).fn);
+  Info[Core::OPINDEX_F80FYL2X] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FYL2X>::handle, Core::OPINDEX_F80FYL2X).fn);
+  Info[Core::OPINDEX_F80ATAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ATAN>::handle, Core::OPINDEX_F80ATAN).fn);
+  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);
+}
+
+bool InterpreterOps::GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info) {
+  uint8_t OpSize = IROp->Size;
+  switch(IROp->Op) {
+    case IR::OP_F80LOADFCW: {
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle, Core::OPINDEX_F80LOADFCW);
+      return true;
+    }
+
+    case IR::OP_F80CVTTO: {
+      auto Op = IROp->C<IR::IROp_F80CVTTo>();
+
+      switch (Op->SrcSize) {
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle4, Core::OPINDEX_F80CVTTO_4);
+          return true;
+        }
+        case 8: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle8, Core::OPINDEX_F80CVTTO_8);
+          return true;
+        }
+      default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CVT: {
+      switch (OpSize) {
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle4, Core::OPINDEX_F80CVT_4);
+          return true;
+        }
+        case 8: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle8, Core::OPINDEX_F80CVT_8);
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CVTINT: {
+      auto Op = IROp->C<IR::IROp_F80CVTInt>();
+
+      switch (OpSize) {
+        case 2: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t, Core::OPINDEX_F80CVTINT_TRUNC2);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2, Core::OPINDEX_F80CVTINT_2);
+          }
+          return true;
+        }
+        case 4: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t, Core::OPINDEX_F80CVTINT_TRUNC4);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4, Core::OPINDEX_F80CVTINT_4);
+          }
+          return true;
+        }
+        case 8: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t, Core::OPINDEX_F80CVTINT_TRUNC8);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8, Core::OPINDEX_F80CVTINT_8);
+          }
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CMP: {
+      auto Op = IROp->C<IR::IROp_F80Cmp>();
+
+      static constexpr std::array handlers{
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<2>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<3>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<4>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<5>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<6>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<7>,
+      };
+
+      *Info = GetFallbackInfo(handlers[Op->Flags], (Core::FallbackHandlerIndex)(Core::OPINDEX_F80CMP_0 + Op->Flags));
+      return true;
+    }
+
+    case IR::OP_F80CVTTOINT: {
+      auto Op = IROp->C<IR::IROp_F80CVTToInt>();
+
+      switch (Op->SrcSize) {
+        case 2: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle2, Core::OPINDEX_F80CVTTOINT_2);
+          return true;
+        }
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle4, Core::OPINDEX_F80CVTTOINT_4);
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+
+#define COMMON_X87_OP(OP) \
+    case IR::OP_F80##OP: { \
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80##OP>::handle, Core::OPINDEX_F80##OP); \
+      return true; \
+    }
+
+    // Unary
+    COMMON_X87_OP(ROUND)
+    COMMON_X87_OP(F2XM1)
+    COMMON_X87_OP(TAN)
+    COMMON_X87_OP(SQRT)
+    COMMON_X87_OP(SIN)
+    COMMON_X87_OP(COS)
+    COMMON_X87_OP(XTRACT_EXP)
+    COMMON_X87_OP(XTRACT_SIG)
+    COMMON_X87_OP(BCDSTORE)
+    COMMON_X87_OP(BCDLOAD)
+
+    // Binary
+    COMMON_X87_OP(ADD)
+    COMMON_X87_OP(SUB)
+    COMMON_X87_OP(MUL)
+    COMMON_X87_OP(DIV)
+    COMMON_X87_OP(FYL2X)
+    COMMON_X87_OP(ATAN)
+    COMMON_X87_OP(FPREM1)
+    COMMON_X87_OP(FPREM)
+    COMMON_X87_OP(SCALE)
+
+    default:
+      break;
+  }
+
+  return false;
+}
+
+
+}

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

@@ -114,7 +114,6 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   // Branch ops
   REGISTER_OP(GUESTCALLDIRECT,        GuestCallDirect);
   REGISTER_OP(GUESTCALLINDIRECT,      GuestCallIndirect);
-  REGISTER_OP(GUESTRETURN,            GuestReturn);
   REGISTER_OP(SIGNALRETURN,           SignalReturn);
   REGISTER_OP(CALLBACKRETURN,         CallbackReturn);
   REGISTER_OP(EXITFUNCTION,           ExitFunction);
@@ -176,6 +175,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(SETROUNDINGMODE,        SetRoundingMode);
   REGISTER_OP(INVALIDATEFLAGS,        NoOp);
   REGISTER_OP(PROCESSORID,            ProcessorID);
+  REGISTER_OP(RDRAND,                 RDRAND);
 
   // Move ops
   REGISTER_OP(EXTRACTELEMENTPAIR,     ExtractElementPair);
@@ -185,8 +185,6 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   // Vector ops
   REGISTER_OP(VECTORZERO,             VectorZero);
   REGISTER_OP(VECTORIMM,              VectorImm);
-  REGISTER_OP(CREATEVECTOR2,          CreateVector2);
-  REGISTER_OP(CREATEVECTOR4,          CreateVector4);
   REGISTER_OP(SPLATVECTOR2,           SplatVector);
   REGISTER_OP(SPLATVECTOR4,           SplatVector);
   REGISTER_OP(VMOV,                   VMov);
@@ -275,6 +273,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(VSMULL2,                VSMull2);
   REGISTER_OP(VUABDL,                 VUABDL);
   REGISTER_OP(VTBL1,                  VTBL1);
+  REGISTER_OP(VREV64,                 VRev64);
 
   // Encryption ops
   REGISTER_OP(VAESIMC,                AESImc);
@@ -283,6 +282,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(VAESDEC,                AESDec);
   REGISTER_OP(VAESDECLAST,            AESDecLast);
   REGISTER_OP(VAESKEYGENASSIST,       AESKeyGenAssist);
+  REGISTER_OP(CRC32,                  CRC32);
 
   // F80 ops
   REGISTER_OP(F80LOADFCW,             F80LOADFCW);
@@ -321,214 +321,9 @@ 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) {
 }
 
-template<typename R, typename... Args>
-static FallbackInfo GetFallbackInfo(R(*fn)(Args...)) {
-  return {FABI_UNKNOWN, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(float)) {
-  return {FABI_F80_F32, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(double)) {
-  return {FABI_F80_F64, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int16_t)) {
-  return {FABI_F80_I16, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(void(*fn)(uint16_t)) {
-  return {FABI_VOID_U16, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int32_t)) {
-  return {FABI_F80_I32, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(float(*fn)(X80SoftFloat)) {
-  return {FABI_F32_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(double(*fn)(X80SoftFloat)) {
-  return {FABI_F64_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(int16_t(*fn)(X80SoftFloat)) {
-  return {FABI_I16_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(int32_t(*fn)(X80SoftFloat)) {
-  return {FABI_I32_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(int64_t(*fn)(X80SoftFloat)) {
-  return {FABI_I64_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(uint64_t(*fn)(X80SoftFloat, X80SoftFloat)) {
-  return {FABI_I64_F80_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat)) {
-  return {FABI_F80_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat, X80SoftFloat)) {
-  return {FABI_F80_F80_F80, (void*)fn};
-}
-
-bool InterpreterOps::GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info) {
-  uint8_t OpSize = IROp->Size;
-  switch(IROp->Op) {
-    case IR::OP_F80LOADFCW: {
-      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle);
-      return true;
-    }
-
-    case IR::OP_F80CVTTO: {
-      auto Op = IROp->C<IR::IROp_F80CVTTo>();
-
-      switch (Op->Size) {
-        case 4: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle4);
-          return true;
-        }
-        case 8: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle8);
-          return true;
-        }
-      default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
-      }
-      break;
-    }
-    case IR::OP_F80CVT: {
-      switch (OpSize) {
-        case 4: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle4);
-          return true;
-        }
-        case 8: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle8);
-          return true;
-        }
-        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
-      }
-      break;
-    }
-    case IR::OP_F80CVTINT: {
-      auto Op = IROp->C<IR::IROp_F80CVTInt>();
-
-      switch (OpSize) {
-        case 2: {
-          *Info = GetFallbackInfo(Op->Truncate ? &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t : &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2);
-          return true;
-        }
-        case 4: {
-          *Info = GetFallbackInfo(Op->Truncate ? &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t : &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4);
-          return true;
-        }
-        case 8: {
-          *Info = GetFallbackInfo(Op->Truncate ? &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t : &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8);
-          return true;
-        }
-        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
-      }
-      break;
-    }
-    case IR::OP_F80CMP: {
-      auto Op = IROp->C<IR::IROp_F80Cmp>();
-
-      static constexpr std::array handlers{
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<2>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<3>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<4>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<5>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<6>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<7>,
-      };
-
-      *Info = GetFallbackInfo(handlers[Op->Flags]);
-      return true;
-    }
-
-    case IR::OP_F80CVTTOINT: {
-      auto Op = IROp->C<IR::IROp_F80CVTToInt>();
-
-      switch (Op->Size) {
-        case 2: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle2);
-          return true;
-        }
-        case 4: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle4);
-          return true;
-        }
-        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
-      }
-      break;
-    }
-
-#define COMMON_X87_OP(OP) \
-    case IR::OP_F80##OP: { \
-      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80##OP>::handle); \
-      return true; \
-    }
-
-    // Unary
-    COMMON_X87_OP(ROUND)
-    COMMON_X87_OP(F2XM1)
-    COMMON_X87_OP(TAN)
-    COMMON_X87_OP(SQRT)
-    COMMON_X87_OP(SIN)
-    COMMON_X87_OP(COS)
-    COMMON_X87_OP(XTRACT_EXP)
-    COMMON_X87_OP(XTRACT_SIG)
-    COMMON_X87_OP(BCDSTORE)
-    COMMON_X87_OP(BCDLOAD)
-
-    // Binary
-    COMMON_X87_OP(ADD)
-    COMMON_X87_OP(SUB)
-    COMMON_X87_OP(MUL)
-    COMMON_X87_OP(DIV)
-    COMMON_X87_OP(FYL2X)
-    COMMON_X87_OP(ATAN)
-    COMMON_X87_OP(FPREM1)
-    COMMON_X87_OP(FPREM)
-    COMMON_X87_OP(SCALE)
-
-    default:
-      break;
-  }
-
-  return false;
-}
-
 void InterpreterOps::InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData) {
   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);

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

@@ -1,6 +1,7 @@
 #pragma once
 #include <stdint.h>
 
+#include <FEXCore/Core/CoreState.h>
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 
@@ -38,12 +39,14 @@ namespace FEXCore::CPU {
   struct FallbackInfo {
     FallbackABI ABI;
     void *fn;
+    FEXCore::Core::FallbackHandlerIndex HandlerIndex;
   };
 
   class InterpreterOps {
 
     public:
       static void InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData);
+      static void FillFallbackIndexPointers(uint64_t *Info);
       static bool GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info);
 
       struct IROpData {
@@ -139,7 +142,6 @@ namespace FEXCore::CPU {
   ///< Branch ops
   DEF_OP(GuestCallDirect);
   DEF_OP(GuestCallIndirect);
-  DEF_OP(GuestReturn);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -194,6 +196,7 @@ namespace FEXCore::CPU {
   DEF_OP(GetRoundingMode);
   DEF_OP(SetRoundingMode);
   DEF_OP(ProcessorID);
+  DEF_OP(RDRAND);
 
   ///< Move ops
   DEF_OP(ExtractElementPair);
@@ -203,8 +206,6 @@ namespace FEXCore::CPU {
   ///< Vector ops
   DEF_OP(VectorZero);
   DEF_OP(VectorImm);
-  DEF_OP(CreateVector2);
-  DEF_OP(CreateVector4);
   DEF_OP(SplatVector);
   DEF_OP(VMov);
   DEF_OP(VAnd);
@@ -290,6 +291,7 @@ namespace FEXCore::CPU {
   DEF_OP(VSMull2);
   DEF_OP(VUABDL);
   DEF_OP(VTBL1);
+  DEF_OP(VRev64);
 
   ///< Encryption ops
   DEF_OP(AESImc);
@@ -298,6 +300,7 @@ namespace FEXCore::CPU {
   DEF_OP(AESDec);
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
+  DEF_OP(CRC32);
 
   ///< F80 ops
   DEF_OP(F80LOADFCW);

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

@@ -14,6 +14,7 @@ $end_info$
 #ifdef _M_X86_64
 #include <xmmintrin.h>
 #endif
+#include <sys/random.h>
 
 namespace FEXCore::CPU {
 [[noreturn]]
@@ -148,6 +149,14 @@ DEF_OP(ProcessorID) {
   GD = (CPUNode << 12) | CPU;
 }
 
+DEF_OP(RDRAND) {
+  // We are ignoring Op->GetReseeded in the interpreter
+  uint64_t *DstPtr = GetDest<uint64_t*>(Data->SSAData, Node);
+  ssize_t Result = ::getrandom(&DstPtr[0], 8, 0);
+
+  // Second result is if we managed to read a valid random number or not
+  DstPtr[1] = Result == 8 ? 1 : 0;
+}
 #undef DEF_OP
 
 } // namespace FEXCore::CPU

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

@@ -26,8 +26,8 @@ DEF_OP(CreateElementPair) {
 
   uint8_t *Dst = GetDest<uint8_t*>(Data->SSAData, Node);
 
-  memcpy(Dst, Src_Lower, Op->Header.Size);
-  memcpy(Dst + Op->Header.Size, Src_Upper, Op->Header.Size);
+  memcpy(Dst, Src_Lower, IROp->ElementSize);
+  memcpy(Dst + IROp->ElementSize, Src_Upper, IROp->ElementSize);
 }
 
 DEF_OP(Mov) {

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

@@ -7,6 +7,7 @@ $end_info$
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
+#include <FEXCore/Utils/BitUtils.h>
 
 #include <bit>
 #include <cstdint>
@@ -38,39 +39,6 @@ DEF_OP(VectorImm) {
   memcpy(GDP, Tmp, OpSize);
 }
 
-DEF_OP(CreateVector2) {
-  auto Op = IROp->C<IR::IROp_CreateVector2>();
-  uint8_t OpSize = IROp->Size;
-
-  LOGMAN_THROW_A_FMT(OpSize <= 16, "Can't handle a vector of size: {}", OpSize);
-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  uint8_t Tmp[16];
-  uint8_t ElementSize = OpSize / 2;
-  #define CREATE_VECTOR(elementsize, type) \
-    case elementsize: { \
-      auto *Dst_d = reinterpret_cast<type*>(Tmp); \
-      auto *Src1_d = reinterpret_cast<type*>(Src1); \
-      auto *Src2_d = reinterpret_cast<type*>(Src2); \
-      Dst_d[0] = *Src1_d; \
-      Dst_d[1] = *Src2_d; \
-      break; \
-    }
-  switch (ElementSize) {
-    CREATE_VECTOR(1, uint8_t)
-    CREATE_VECTOR(2, uint16_t)
-    CREATE_VECTOR(4, uint32_t)
-    CREATE_VECTOR(8, uint64_t)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize); break;
-  }
-  #undef CREATE_VECTOR
-  memcpy(GDP, Tmp, OpSize);
-}
-
-DEF_OP(CreateVector4) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
 DEF_OP(SplatVector) {
   auto Op = IROp->C<IR::IROp_SplatVector2>();
   uint8_t OpSize = IROp->Size;
@@ -1402,10 +1370,9 @@ DEF_OP(VInsScalarElement) {
 
 DEF_OP(VExtractElement) {
   auto Op = IROp->C<IR::IROp_VExtractElement>();
-  uint8_t OpSize = IROp->Size;
 
   uint32_t SourceSize = GetOpSize(Data->CurrentIR, Op->Header.Args[0]);
-  LOGMAN_THROW_A_FMT(OpSize <= 16, "OpSize is too large for VExtractElement: {}", OpSize);
+  LOGMAN_THROW_A_FMT(IROp->Size <= 16, "OpSize is too large for VExtractElement: {}", IROp->Size);
   if (SourceSize == 16) {
     __uint128_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
     uint64_t Shift = Op->Header.ElementSize * Op->Index * 8;
@@ -1931,6 +1898,39 @@ DEF_OP(VTBL1) {
   memcpy(GDP, Tmp, OpSize);
 }
 
+DEF_OP(VRev64) {
+  auto Op = IROp->C<IR::IROp_VRev64>();
+  uint8_t OpSize = IROp->Size;
+
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
+
+  uint8_t Tmp[16];
+
+  uint8_t Elements = OpSize / 8;
+
+  // The element working size is always 64-bit
+  // The defined element size in the op is the operating size of the element swapping
+  auto Func8 = [](auto a) { return BSwap64(a); };
+  auto Func16 = [](auto a) {
+    return (a >> 48) |                    // Element[3] -> Element[0]
+      ((a >> 16) & 0xFFFF'0000U) |        // Element[2] -> Element[1]
+      ((a << 16) & 0xFFFF'0000'0000ULL) | // Element[1] -> Element[2]
+      (a << 48);                          // Element[0] -> Element[3]
+  };
+  auto Func32 = [](auto a) {
+    return (a >> 32) | (a << 32);
+  };
+
+  switch (Op->Header.ElementSize) {
+    DO_VECTOR_1SRC_OP(1, uint64_t, Func8)
+    DO_VECTOR_1SRC_OP(2, uint64_t, Func16)
+    DO_VECTOR_1SRC_OP(4, uint64_t, Func32)
+
+    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+  }
+  memcpy(GDP, Tmp, Op->Header.Size);
+}
+
 #undef DEF_OP
 
 } // namespace FEXCore::CPU

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

@@ -12,37 +12,13 @@ namespace FEXCore::CPU {
 #define GRD(Node) (IROp->Size <= 4 ? GetDst<RA_32>(Node) : GetDst<RA_64>(Node))
 #define GRS(Node) (IROp->Size <= 4 ? GetReg<RA_32>(Node) : GetReg<RA_64>(Node))
 
-static uint64_t LUDIV(uint64_t SrcHigh, uint64_t SrcLow, uint64_t Divisor) {
-  __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
-  __uint128_t Res = Source / Divisor;
-  return Res;
-}
-
-static int64_t LDIV(int64_t SrcHigh, int64_t SrcLow, int64_t Divisor) {
-  __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
-  __int128_t Res = Source / Divisor;
-  return Res;
-}
-
-static uint64_t LUREM(uint64_t SrcHigh, uint64_t SrcLow, uint64_t Divisor) {
-  __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
-  __uint128_t Res = Source % Divisor;
-  return Res;
-}
-
-static int64_t LREM(int64_t SrcHigh, int64_t SrcLow, int64_t Divisor) {
-  __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
-  __int128_t Res = Source % Divisor;
-  return Res;
-}
-
 using namespace vixl;
 using namespace vixl::aarch64;
 #define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 DEF_OP(TruncElementPair) {
   auto Op = IROp->C<IR::IROp_TruncElementPair>();
 
-  switch (Op->Size) {
+  switch (IROp->Size) {
     case 4: {
       auto Dst = GetSrcPair<RA_32>(Node);
       auto Src = GetSrcPair<RA_32>(Op->Header.Args[0].ID());
@@ -50,7 +26,7 @@ DEF_OP(TruncElementPair) {
       mov(Dst.second, Src.second);
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unhandled Truncation size: {}", Op->Size); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled Truncation size: {}", IROp->Size); break;
   }
 }
 
@@ -65,7 +41,13 @@ DEF_OP(EntrypointOffset) {
 
   auto Constant = Entry + Op->Offset;
   auto Dst = GetReg<RA_64>(Node);
-  LoadConstant(Dst, Constant);
+  uint64_t Mask = ~0ULL;
+  uint8_t OpSize = IROp->Size;
+  if (OpSize == 4) {
+    Mask = 0xFFFF'FFFFULL;
+  }
+
+  LoadConstant(Dst, Constant & Mask);
 }
 
 DEF_OP(InlineConstant) {
@@ -654,23 +636,42 @@ DEF_OP(LDiv) {
     break;
     }
     case 8: {
-      PushDynamicRegsAndLR();
+      auto Upper64Bit = GetReg<RA_64>(Op->Header.Args[1].ID());
+      auto Lower64Bit = GetReg<RA_64>(Op->Header.Args[0].ID());
+      auto Divisor = GetReg<RA_64>(Op->Header.Args[2].ID());
+      Label Only64Bit{};
+      Label LongDIVRet{};
 
-      mov(x0, GetReg<RA_64>(Op->Header.Args[1].ID()));
-      mov(x1, GetReg<RA_64>(Op->Header.Args[0].ID()));
-      mov(x2, GetReg<RA_64>(Op->Header.Args[2].ID()));
+      // Check if the upper bits match the top bit of the lower 64-bits
+      // Sign extend the top bit of lower bits
+      sbfx(TMP1, Lower64Bit, 63, 1);
+      eor(TMP1, TMP1, Upper64Bit);
 
-      LoadConstant(x3, reinterpret_cast<uint64_t>(LDIV));
-      SpillStaticRegs();
-      blr(x3);
-      FillStaticRegs();
+      // If the sign bit matches then the result is zero
+      cbz(TMP1, &Only64Bit);
 
-      // Result is now in x0
-      // Fix the stack and any values that were stepped on
-      PopDynamicRegsAndLR();
+      // Long divide
+      {
+        mov(x0, Upper64Bit);
+        mov(x1, Lower64Bit);
+        mov(x2, Divisor);
 
-      // Move result to its destination register
-      mov(GetReg<RA_64>(Node), x0);
+        ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LDIVHandler)));
+        blr(x3);
+        // Move result to its destination register
+        mov(GetReg<RA_64>(Node), x0);
+
+        // Skip 64-bit path
+        b(&LongDIVRet);
+      }
+
+      bind(&Only64Bit);
+      // 64-Bit only
+      {
+        sdiv(GetReg<RA_64>(Node), Lower64Bit, Divisor);
+      }
+
+      bind(&LongDIVRet);
     break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown LDIV Size: {}", Size); break;
@@ -697,23 +698,38 @@ DEF_OP(LUDiv) {
     break;
     }
     case 8: {
-      PushDynamicRegsAndLR();
+      auto Upper64Bit = GetReg<RA_64>(Op->Header.Args[1].ID());
+      auto Lower64Bit = GetReg<RA_64>(Op->Header.Args[0].ID());
+      auto Divisor = GetReg<RA_64>(Op->Header.Args[2].ID());
+      Label Only64Bit{};
+      Label LongDIVRet{};
 
-      mov(x0, GetReg<RA_64>(Op->Header.Args[1].ID()));
-      mov(x1, GetReg<RA_64>(Op->Header.Args[0].ID()));
-      mov(x2, GetReg<RA_64>(Op->Header.Args[2].ID()));
+      // Check the upper bits for zero
+      // If the upper bits are zero then we can do a 64-bit divide
+      cbz(Upper64Bit, &Only64Bit);
 
-      LoadConstant(x3, reinterpret_cast<uint64_t>(LUDIV));
-      SpillStaticRegs();
-      blr(x3);
-      FillStaticRegs();
+      // Long divide
+      {
+        mov(x0, Upper64Bit);
+        mov(x1, Lower64Bit);
+        mov(x2, Divisor);
 
-      // Result is now in x0
-      // Fix the stack and any values that were stepped on
-      PopDynamicRegsAndLR();
+        ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LUDIVHandler)));
+        blr(x3);
+        // Move result to its destination register
+        mov(GetReg<RA_64>(Node), x0);
 
-      // Move result to its destination register
-      mov(GetReg<RA_64>(Node), x0);
+        // Skip 64-bit path
+        b(&LongDIVRet);
+      }
+
+      bind(&Only64Bit);
+      // 64-Bit only
+      {
+        udiv(GetReg<RA_64>(Node), Lower64Bit, Divisor);
+      }
+
+      bind(&LongDIVRet);
     break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown LUDIV Size: {}", Size); break;
@@ -750,23 +766,42 @@ DEF_OP(LRem) {
     break;
     }
     case 8: {
-      PushDynamicRegsAndLR();
+      auto Upper64Bit = GetReg<RA_64>(Op->Header.Args[1].ID());
+      auto Lower64Bit = GetReg<RA_64>(Op->Header.Args[0].ID());
+      auto Divisor = GetReg<RA_64>(Op->Header.Args[2].ID());
+      Label Only64Bit{};
+      Label LongDIVRet{};
 
-      mov(x0, GetReg<RA_64>(Op->Header.Args[1].ID()));
-      mov(x1, GetReg<RA_64>(Op->Header.Args[0].ID()));
-      mov(x2, GetReg<RA_64>(Op->Header.Args[2].ID()));
+      // Check if the upper bits match the top bit of the lower 64-bits
+      // Sign extend the top bit of lower bits
+      sbfx(TMP1, Lower64Bit, 63, 1);
+      eor(TMP1, TMP1, Upper64Bit);
 
-      LoadConstant(x3, reinterpret_cast<uint64_t>(LREM));
-      SpillStaticRegs();
-      blr(x3);
-      FillStaticRegs();
+      // If the sign bit matches then the result is zero
+      cbz(TMP1, &Only64Bit);
 
-      // Result is now in x0
-      // Fix the stack and any values that were stepped on
-      PopDynamicRegsAndLR();
+      // Long divide
+      {
+        mov(x0, Upper64Bit);
+        mov(x1, Lower64Bit);
+        mov(x2, Divisor);
 
-      // Move result to its destination register
-      mov(GetReg<RA_64>(Node), x0);
+        ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LREMHandler)));
+        blr(x3);
+        // Move result to its destination register
+        mov(GetReg<RA_64>(Node), x0);
+
+        // Skip 64-bit path
+        b(&LongDIVRet);
+      }
+
+      bind(&Only64Bit);
+      // 64-Bit only
+      {
+        sdiv(TMP1, Lower64Bit, Divisor);
+        msub(GetReg<RA_64>(Node), TMP1, Divisor, Lower64Bit);
+      }
+      bind(&LongDIVRet);
     break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown LREM Size: {}", Size); break;
@@ -799,24 +834,39 @@ DEF_OP(LURem) {
     break;
     }
     case 8: {
+      auto Upper64Bit = GetReg<RA_64>(Op->Header.Args[1].ID());
+      auto Lower64Bit = GetReg<RA_64>(Op->Header.Args[0].ID());
+      auto Divisor = GetReg<RA_64>(Op->Header.Args[2].ID());
+      Label Only64Bit{};
+      Label LongDIVRet{};
 
-      PushDynamicRegsAndLR();
+      // Check the upper bits for zero
+      // If the upper bits are zero then we can do a 64-bit divide
+      cbz(Upper64Bit, &Only64Bit);
 
-      mov(x0, GetReg<RA_64>(Op->Header.Args[1].ID()));
-      mov(x1, GetReg<RA_64>(Op->Header.Args[0].ID()));
-      mov(x2, GetReg<RA_64>(Op->Header.Args[2].ID()));
+      // Long divide
+      {
+        mov(x0, Upper64Bit);
+        mov(x1, Lower64Bit);
+        mov(x2, Divisor);
 
-      LoadConstant(x3, reinterpret_cast<uint64_t>(LUREM));
-      SpillStaticRegs();
-      blr(x3);
-      FillStaticRegs();
+        ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LUREMHandler)));
+        blr(x3);
+        // Move result to its destination register
+        mov(GetReg<RA_64>(Node), x0);
 
-      // Fix the stack and any values that were stepped on
-      PopDynamicRegsAndLR();
+        // Skip 64-bit path
+        b(&LongDIVRet);
+      }
 
-      // Result is now in x0
-      // Move result to its destination register
-      mov(GetReg<RA_64>(Node), x0);
+      bind(&Only64Bit);
+      // 64-Bit only
+      {
+        udiv(TMP1, Lower64Bit, Divisor);
+        msub(GetReg<RA_64>(Node), TMP1, Divisor, Lower64Bit);
+      }
+
+      bind(&LongDIVRet);
     break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown LUREM Size: {}", OpSize); break;
@@ -1087,16 +1137,16 @@ DEF_OP(VExtractToGPR) {
   uint8_t OpSize = IROp->Size;
   switch (OpSize) {
     case 1:
-      umov(GetReg<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()).V16B(), Op->Idx);
+      umov(GetReg<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()).V16B(), Op->Index);
     break;
     case 2:
-      umov(GetReg<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()).V8H(), Op->Idx);
+      umov(GetReg<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()).V8H(), Op->Index);
     break;
     case 4:
-      umov(GetReg<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()).V4S(), Op->Idx);
+      umov(GetReg<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()).V4S(), Op->Index);
     break;
     case 8:
-      umov(GetReg<RA_64>(Node), GetSrc(Op->Header.Args[0].ID()).V2D(), Op->Idx);
+      umov(GetReg<RA_64>(Node), GetSrc(Op->Header.Args[0].ID()).V2D(), Op->Index);
     break;
     default:  LOGMAN_MSG_A_FMT("Unhandled ExtractElementSize: {}", OpSize);
   }

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

@@ -12,18 +12,17 @@ using namespace vixl::aarch64;
 #define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 DEF_OP(CASPair) {
   auto Op = IROp->C<IR::IROp_CASPair>();
-  uint8_t OpSize = IROp->Size;
   // Size is the size of each pair element
   auto Dst = GetSrcPair<RA_64>(Node);
-  auto Expected = GetSrcPair<RA_64>(Op->Header.Args[0].ID());
-  auto Desired = GetSrcPair<RA_64>(Op->Header.Args[1].ID());
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[2].ID());
+  auto Expected = GetSrcPair<RA_64>(Op->Expected.ID());
+  auto Desired = GetSrcPair<RA_64>(Op->Desired.ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
     mov(TMP3, Expected.first);
     mov(TMP4, Expected.second);
 
-    switch (OpSize) {
+    switch (IROp->ElementSize) {
     case 4:
       caspal(TMP3.W(), TMP4.W(), Desired.first.W(), Desired.second.W(), MemOperand(MemSrc));
       mov(Dst.first.W(), TMP3.W());
@@ -34,11 +33,11 @@ DEF_OP(CASPair) {
       mov(Dst.first, TMP3);
       mov(Dst.second, TMP4);
       break;
-    default: LOGMAN_MSG_A_FMT("Unsupported: {}", OpSize);
+    default: LOGMAN_MSG_A_FMT("Unsupported: {}", IROp->ElementSize);
     }
   }
   else {
-    switch (OpSize) {
+    switch (IROp->ElementSize) {
       case 4: {
         aarch64::Label LoopTop;
         aarch64::Label LoopNotExpected;
@@ -91,7 +90,7 @@ DEF_OP(CASPair) {
         bind(&LoopExpected);
         break;
       }
-      default: LOGMAN_MSG_A_FMT("Unsupported: {}", OpSize);
+      default: LOGMAN_MSG_A_FMT("Unsupported: {}", IROp->ElementSize);
     }
   }
 }
@@ -99,16 +98,13 @@ DEF_OP(CASPair) {
 DEF_OP(CAS) {
   auto Op = IROp->C<IR::IROp_CAS>();
   uint8_t OpSize = IROp->Size;
-  // Args[0]: Expected
-  // Args[1]: Desired
-  // Args[2]: Pointer
   // DataSrc = *Src1
   // if (DataSrc == Src3) { *Src1 == Src2; } Src2 = DataSrc
   // This will write to memory! Careful!
 
-  auto Expected = GetReg<RA_64>(Op->Header.Args[0].ID());
-  auto Desired = GetReg<RA_64>(Op->Header.Args[1].ID());
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[2].ID());
+  auto Expected = GetReg<RA_64>(Op->Expected.ID());
+  auto Desired = GetReg<RA_64>(Op->Desired.ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
     mov(TMP2, Expected);
@@ -216,14 +212,14 @@ DEF_OP(CAS) {
 DEF_OP(AtomicAdd) {
   auto Op = IROp->C<IR::IROp_AtomicAdd>();
 
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
     switch (IROp->Size) {
-    case 1: staddlb(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 2: staddlh(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 4: staddl(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 8: staddl(GetReg<RA_64>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
+    case 1: staddlb(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 2: staddlh(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 4: staddl(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 8: staddl(GetReg<RA_64>(Op->Value.ID()), MemOperand(MemSrc)); break;
     default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
     }
   }
@@ -234,7 +230,7 @@ DEF_OP(AtomicAdd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        add(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        add(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -243,7 +239,7 @@ DEF_OP(AtomicAdd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        add(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        add(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -252,7 +248,7 @@ DEF_OP(AtomicAdd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        add(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        add(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -261,7 +257,7 @@ DEF_OP(AtomicAdd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        add(TMP2, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        add(TMP2, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP2, TMP2, MemOperand(MemSrc));
         cbnz(TMP2, &LoopTop);
         break;
@@ -274,10 +270,10 @@ DEF_OP(AtomicAdd) {
 DEF_OP(AtomicSub) {
   auto Op = IROp->C<IR::IROp_AtomicSub>();
 
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
-    neg(TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+    neg(TMP2, GetReg<RA_64>(Op->Value.ID()));
     switch (IROp->Size) {
     case 1: staddlb(TMP2.W(), MemOperand(MemSrc)); break;
     case 2: staddlh(TMP2.W(), MemOperand(MemSrc)); break;
@@ -293,7 +289,7 @@ DEF_OP(AtomicSub) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        sub(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        sub(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -302,7 +298,7 @@ DEF_OP(AtomicSub) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        sub(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        sub(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -311,7 +307,7 @@ DEF_OP(AtomicSub) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        sub(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        sub(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -320,7 +316,7 @@ DEF_OP(AtomicSub) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        sub(TMP2, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        sub(TMP2, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP2, TMP2, MemOperand(MemSrc));
         cbnz(TMP2, &LoopTop);
         break;
@@ -333,10 +329,10 @@ DEF_OP(AtomicSub) {
 DEF_OP(AtomicAnd) {
   auto Op = IROp->C<IR::IROp_AtomicAnd>();
 
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
-    mvn(TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+    mvn(TMP2, GetReg<RA_64>(Op->Value.ID()));
     switch (IROp->Size) {
     case 1: stclrlb(TMP2.W(), MemOperand(MemSrc)); break;
     case 2: stclrlh(TMP2.W(), MemOperand(MemSrc)); break;
@@ -352,7 +348,7 @@ DEF_OP(AtomicAnd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        and_(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        and_(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -361,7 +357,7 @@ DEF_OP(AtomicAnd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        and_(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        and_(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -370,7 +366,7 @@ DEF_OP(AtomicAnd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        and_(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        and_(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -379,7 +375,7 @@ DEF_OP(AtomicAnd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        and_(TMP2, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        and_(TMP2, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP2, TMP2, MemOperand(MemSrc));
         cbnz(TMP2, &LoopTop);
         break;
@@ -392,14 +388,14 @@ DEF_OP(AtomicAnd) {
 DEF_OP(AtomicOr) {
   auto Op = IROp->C<IR::IROp_AtomicOr>();
 
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
     switch (IROp->Size) {
-    case 1: stsetlb(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 2: stsetlh(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 4: stsetl(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 8: stsetl(GetReg<RA_64>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
+    case 1: stsetlb(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 2: stsetlh(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 4: stsetl(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 8: stsetl(GetReg<RA_64>(Op->Value.ID()), MemOperand(MemSrc)); break;
     default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
     }
   }
@@ -410,7 +406,7 @@ DEF_OP(AtomicOr) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        orr(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        orr(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -419,7 +415,7 @@ DEF_OP(AtomicOr) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        orr(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        orr(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -428,7 +424,7 @@ DEF_OP(AtomicOr) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        orr(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        orr(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -437,7 +433,7 @@ DEF_OP(AtomicOr) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        orr(TMP2, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        orr(TMP2, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP2, TMP2, MemOperand(MemSrc));
         cbnz(TMP2, &LoopTop);
         break;
@@ -450,14 +446,14 @@ DEF_OP(AtomicOr) {
 DEF_OP(AtomicXor) {
   auto Op = IROp->C<IR::IROp_AtomicXor>();
 
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
     switch (IROp->Size) {
-    case 1: steorlb(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 2: steorlh(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 4: steorl(GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
-    case 8: steorl(GetReg<RA_64>(Op->Header.Args[1].ID()), MemOperand(MemSrc)); break;
+    case 1: steorlb(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 2: steorlh(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 4: steorl(GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc)); break;
+    case 8: steorl(GetReg<RA_64>(Op->Value.ID()), MemOperand(MemSrc)); break;
     default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
     }
   }
@@ -468,7 +464,7 @@ DEF_OP(AtomicXor) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        eor(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        eor(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -477,7 +473,7 @@ DEF_OP(AtomicXor) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        eor(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        eor(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -486,7 +482,7 @@ DEF_OP(AtomicXor) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        eor(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        eor(TMP2.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP2.W(), TMP2.W(), MemOperand(MemSrc));
         cbnz(TMP2.W(), &LoopTop);
         break;
@@ -495,7 +491,7 @@ DEF_OP(AtomicXor) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        eor(TMP2, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        eor(TMP2, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP2, TMP2, MemOperand(MemSrc));
         cbnz(TMP2, &LoopTop);
         break;
@@ -508,10 +504,10 @@ DEF_OP(AtomicXor) {
 DEF_OP(AtomicSwap) {
   auto Op = IROp->C<IR::IROp_AtomicSwap>();
 
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
-    mov(TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+    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;
@@ -527,7 +523,7 @@ DEF_OP(AtomicSwap) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        stlxrb(TMP4.W(), GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc));
+        stlxrb(TMP4.W(), GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         uxtb(GetReg<RA_32>(Node), TMP2.W());
         break;
@@ -536,7 +532,7 @@ DEF_OP(AtomicSwap) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        stlxrh(TMP4.W(), GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc));
+        stlxrh(TMP4.W(), GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         uxtw(GetReg<RA_32>(Node), TMP2.W());
         break;
@@ -545,7 +541,7 @@ DEF_OP(AtomicSwap) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        stlxr(TMP4.W(), GetReg<RA_32>(Op->Header.Args[1].ID()), MemOperand(MemSrc));
+        stlxr(TMP4.W(), GetReg<RA_32>(Op->Value.ID()), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
         break;
@@ -554,7 +550,7 @@ DEF_OP(AtomicSwap) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        stlxr(TMP4, GetReg<RA_64>(Op->Header.Args[1].ID()), MemOperand(MemSrc));
+        stlxr(TMP4, GetReg<RA_64>(Op->Value.ID()), MemOperand(MemSrc));
         cbnz(TMP4, &LoopTop);
         mov(GetReg<RA_64>(Node), TMP2.X());
         break;
@@ -566,14 +562,14 @@ DEF_OP(AtomicSwap) {
 
 DEF_OP(AtomicFetchAdd) {
   auto Op = IROp->C<IR::IROp_AtomicFetchAdd>();
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
     switch (IROp->Size) {
-    case 1: ldaddalb(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 2: ldaddalh(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 4: ldaddal(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 8: ldaddal(GetReg<RA_64>(Op->Header.Args[1].ID()), GetReg<RA_64>(Node), MemOperand(MemSrc)); break;
+    case 1: ldaddalb(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 2: ldaddalh(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 4: ldaddal(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 8: ldaddal(GetReg<RA_64>(Op->Value.ID()), GetReg<RA_64>(Node), MemOperand(MemSrc)); break;
     default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
     }
   }
@@ -584,7 +580,7 @@ DEF_OP(AtomicFetchAdd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        add(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        add(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -594,7 +590,7 @@ DEF_OP(AtomicFetchAdd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        add(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        add(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -604,7 +600,7 @@ DEF_OP(AtomicFetchAdd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        add(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        add(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -614,7 +610,7 @@ DEF_OP(AtomicFetchAdd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        add(TMP3, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        add(TMP3, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP4, TMP3, MemOperand(MemSrc));
         cbnz(TMP4, &LoopTop);
         mov(GetReg<RA_64>(Node), TMP2);
@@ -627,10 +623,10 @@ DEF_OP(AtomicFetchAdd) {
 
 DEF_OP(AtomicFetchSub) {
   auto Op = IROp->C<IR::IROp_AtomicFetchSub>();
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
-    neg(TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+    neg(TMP2, GetReg<RA_64>(Op->Value.ID()));
     switch (IROp->Size) {
     case 1: ldaddalb(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
     case 2: ldaddalh(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
@@ -646,7 +642,7 @@ DEF_OP(AtomicFetchSub) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        sub(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        sub(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -656,7 +652,7 @@ DEF_OP(AtomicFetchSub) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        sub(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        sub(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -666,7 +662,7 @@ DEF_OP(AtomicFetchSub) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        sub(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        sub(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -676,7 +672,7 @@ DEF_OP(AtomicFetchSub) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        sub(TMP3, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        sub(TMP3, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP4, TMP3, MemOperand(MemSrc));
         cbnz(TMP4, &LoopTop);
         mov(GetReg<RA_64>(Node), TMP2);
@@ -689,10 +685,10 @@ DEF_OP(AtomicFetchSub) {
 
 DEF_OP(AtomicFetchAnd) {
   auto Op = IROp->C<IR::IROp_AtomicFetchAnd>();
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
-    mvn(TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+    mvn(TMP2, GetReg<RA_64>(Op->Value.ID()));
     switch (IROp->Size) {
     case 1: ldclralb(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
     case 2: ldclralh(TMP2.W(), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
@@ -708,7 +704,7 @@ DEF_OP(AtomicFetchAnd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        and_(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        and_(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -718,7 +714,7 @@ DEF_OP(AtomicFetchAnd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        and_(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        and_(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -728,7 +724,7 @@ DEF_OP(AtomicFetchAnd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        and_(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        and_(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -738,7 +734,7 @@ DEF_OP(AtomicFetchAnd) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        and_(TMP3, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        and_(TMP3, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP4, TMP3, MemOperand(MemSrc));
         cbnz(TMP4, &LoopTop);
         mov(GetReg<RA_64>(Node), TMP2);
@@ -751,14 +747,14 @@ DEF_OP(AtomicFetchAnd) {
 
 DEF_OP(AtomicFetchOr) {
   auto Op = IROp->C<IR::IROp_AtomicFetchOr>();
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
     switch (IROp->Size) {
-    case 1: ldsetalb(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 2: ldsetalh(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 4: ldsetal(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 8: ldsetal(GetReg<RA_64>(Op->Header.Args[1].ID()), GetReg<RA_64>(Node), MemOperand(MemSrc)); break;
+    case 1: ldsetalb(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 2: ldsetalh(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 4: ldsetal(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 8: ldsetal(GetReg<RA_64>(Op->Value.ID()), GetReg<RA_64>(Node), MemOperand(MemSrc)); break;
     default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
     }
   }
@@ -769,7 +765,7 @@ DEF_OP(AtomicFetchOr) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        orr(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        orr(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -779,7 +775,7 @@ DEF_OP(AtomicFetchOr) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        orr(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        orr(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -789,7 +785,7 @@ DEF_OP(AtomicFetchOr) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        orr(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        orr(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -799,7 +795,7 @@ DEF_OP(AtomicFetchOr) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        orr(TMP3, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        orr(TMP3, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP4, TMP3, MemOperand(MemSrc));
         cbnz(TMP4, &LoopTop);
         mov(GetReg<RA_64>(Node), TMP2);
@@ -812,14 +808,14 @@ DEF_OP(AtomicFetchOr) {
 
 DEF_OP(AtomicFetchXor) {
   auto Op = IROp->C<IR::IROp_AtomicFetchXor>();
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   if (CTX->HostFeatures.SupportsAtomics) {
     switch (IROp->Size) {
-    case 1: ldeoralb(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 2: ldeoralh(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 4: ldeoral(GetReg<RA_32>(Op->Header.Args[1].ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
-    case 8: ldeoral(GetReg<RA_64>(Op->Header.Args[1].ID()), GetReg<RA_64>(Node), MemOperand(MemSrc)); break;
+    case 1: ldeoralb(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 2: ldeoralh(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 4: ldeoral(GetReg<RA_32>(Op->Value.ID()), GetReg<RA_32>(Node), MemOperand(MemSrc)); break;
+    case 8: ldeoral(GetReg<RA_64>(Op->Value.ID()), GetReg<RA_64>(Node), MemOperand(MemSrc)); break;
     default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
     }
   }
@@ -830,7 +826,7 @@ DEF_OP(AtomicFetchXor) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrb(TMP2.W(), MemOperand(MemSrc));
-        eor(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        eor(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrb(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -840,7 +836,7 @@ DEF_OP(AtomicFetchXor) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxrh(TMP2.W(), MemOperand(MemSrc));
-        eor(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        eor(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxrh(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -850,7 +846,7 @@ DEF_OP(AtomicFetchXor) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2.W(), MemOperand(MemSrc));
-        eor(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Header.Args[1].ID()));
+        eor(TMP3.W(), TMP2.W(), GetReg<RA_32>(Op->Value.ID()));
         stlxr(TMP4.W(), TMP3.W(), MemOperand(MemSrc));
         cbnz(TMP4.W(), &LoopTop);
         mov(GetReg<RA_32>(Node), TMP2.W());
@@ -860,7 +856,7 @@ DEF_OP(AtomicFetchXor) {
         aarch64::Label LoopTop;
         bind(&LoopTop);
         ldaxr(TMP2, MemOperand(MemSrc));
-        eor(TMP3, TMP2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+        eor(TMP3, TMP2, GetReg<RA_64>(Op->Value.ID()));
         stlxr(TMP4, TMP3, MemOperand(MemSrc));
         cbnz(TMP4, &LoopTop);
         mov(GetReg<RA_64>(Node), TMP2);
@@ -873,7 +869,7 @@ DEF_OP(AtomicFetchXor) {
 
 DEF_OP(AtomicFetchNeg) {
   auto Op = IROp->C<IR::IROp_AtomicFetchNeg>();
-  auto MemSrc = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto MemSrc = GetReg<RA_64>(Op->Addr.ID());
 
   // TMP2-TMP3
   switch (IROp->Size) {

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

@@ -4,6 +4,7 @@ tags: backend|arm64
 $end_info$
 */
 
+#include "FEXCore/IR/IR.h"
 #include "Interface/Core/LookupCache.h"
 
 #include "Interface/Core/JIT/Arm64/JITClass.h"
@@ -26,17 +27,13 @@ DEF_OP(GuestCallIndirect) {
   LogMan::Msg::DFmt("Unimplemented");
 }
 
-DEF_OP(GuestReturn) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
 DEF_OP(SignalReturn) {
   // First we must reset the stack
   ResetStack();
 
   // Now branch to our signal return helper
   // This can't be a direct branch since the code needs to live at a constant location
-  LoadConstant(x0, ThreadSharedData.SignalReturnInstruction);
+  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SignalReturnHandler)));
   br(x0);
 }
 
@@ -49,7 +46,7 @@ DEF_OP(CallbackReturn) {
   ResetStack();
 
   // We can now lower the ref counter again
-  LoadConstant(x0, reinterpret_cast<uint64_t>(ThreadSharedData.SignalHandlerRefCounterPtr));
+  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SignalHandlerRefCountPointer)));
   ldr(w2, MemOperand(x0));
   sub(w2, w2, 1);
   str(w2, MemOperand(x0));
@@ -88,7 +85,7 @@ DEF_OP(ExitFunction) {
     RipReg = GetReg<RA_64>(Op->Header.Args[0].ID());
 
     // L1 Cache
-    LoadConstant(x0, ThreadState->LookupCache->GetL1Pointer());
+    ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.L1Pointer)));
 
     and_(x3, RipReg, LookupCache::L1_ENTRIES_MASK);
     add(x0, x0, Operand(x3, Shift::LSL, 4));
@@ -99,7 +96,7 @@ DEF_OP(ExitFunction) {
     br(x1);
 
     bind(&FullLookup);
-    LoadConstant(TMP1, ThreadSharedData.Dispatcher->AbsoluteLoopTopAddress);
+    ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.DispatcherLoopTop)));
     str(RipReg, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
     br(TMP1);
   }
@@ -184,8 +181,16 @@ DEF_OP(Syscall) {
   // X1: ThreadState
   // X2: Pointer to SyscallArguments
 
+  FEXCore::IR::SyscallFlags Flags = Op->Flags;
   PushDynamicRegsAndLR();
-  SpillStaticRegs();
+
+  if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) != FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) {
+    SpillStaticRegs();
+  }
+  else {
+    // Need to spill all caller saved registers still
+    SpillStaticRegs(true, CALLER_GPR_MASK, CALLER_FPR_MASK);
+  }
 
   uint64_t SPOffset = AlignUp(FEXCore::HLE::SyscallArguments::MAX_ARGS * 8, 16);
   sub(sp, sp, SPOffset);
@@ -194,22 +199,30 @@ DEF_OP(Syscall) {
     str(GetReg<RA_64>(Op->Header.Args[i].ID()), MemOperand(sp, i * 8));
   }
 
-  LoadConstant(x0, reinterpret_cast<uint64_t>(CTX->SyscallHandler));
+  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SyscallHandlerObj)));
+  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SyscallHandlerFunc)));
   mov(x1, STATE);
   mov(x2, sp);
-
-  LoadConstant(x3, reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall));
   blr(x3);
 
   add(sp, sp, SPOffset);
 
-  // Result is now in x0
-  // Fix the stack and any values that were stepped on
-  FillStaticRegs();
+  if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) != FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY &&
+      (Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
+    FillStaticRegs();
+  }
+  else {
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    FillStaticRegs(true, CALLER_GPR_MASK, CALLER_FPR_MASK);
+  }
+
   PopDynamicRegsAndLR();
 
-  // Move result to its destination register
-  mov(GetReg<RA_64>(Node), x0);
+  if ((Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
+    // Move result to its destination register
+    mov(GetReg<RA_64>(Node), x0);
+  }
 }
 
 DEF_OP(InlineSyscall) {
@@ -340,21 +353,23 @@ DEF_OP(InlineSyscall) {
   svc(0);
   // On updated signal mask we can receive a signal RIGHT HERE
 
-  // Now that we are done in the syscall we need to carefully peel back the state
-  // First unspill the registers from before
-  FillStaticRegs(false, SpillMask);
+  if ((Op->Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
+    // Now that we are done in the syscall we need to carefully peel back the state
+    // First unspill the registers from before
+    FillStaticRegs(false, SpillMask);
 
-  // Now the registers we've spilled are back in their original host registers
-  // We can safely claim we are no longer in a syscall
-  str(xzr, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
+    // Now the registers we've spilled are back in their original host registers
+    // We can safely claim we are no longer in a syscall
+    str(xzr, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
 
-  // Result is now in x0
-  // Move result to its destination register
-  if (CTX->Config.Is64BitMode()) {
-    mov(GetReg<RA_64>(Node), x0);
-  }
-  else {
-    uxtw(GetReg<RA_64>(Node), x0);
+    // Result is now in x0
+    // Move result to its destination register
+    if (CTX->Config.Is64BitMode()) {
+      mov(GetReg<RA_64>(Node), x0);
+    }
+    else {
+      uxtw(GetReg<RA_64>(Node), x0);
+    }
   }
 }
 
@@ -437,7 +452,7 @@ DEF_OP(RemoveCodeEntry) {
   mov(x0, STATE);
   LoadConstant(x1, Entry);
 
-  LoadConstant(x2, reinterpret_cast<uintptr_t>(&Context::Context::RemoveCodeEntryFromJit));
+  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.RemoveCodeEntryFromJIT)));
   SpillStaticRegs();
   blr(x2);
   FillStaticRegs();
@@ -454,19 +469,10 @@ DEF_OP(CPUID) {
   // x0 = CPUID Handler
   // x1 = CPUID Function
   // x2 = CPUID Leaf
-  LoadConstant(x0, reinterpret_cast<uint64_t>(&CTX->CPUID));
+  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()));
-
-  using ClassPtrType = FEXCore::CPUID::FunctionResults (FEXCore::CPUIDEmu::*)(uint32_t, uint32_t);
-  union PtrCast {
-    ClassPtrType ClassPtr;
-    uintptr_t Data;
-  };
-
-  PtrCast Ptr;
-  Ptr.ClassPtr = &FEXCore::CPUIDEmu::RunFunction;
-  LoadConstant(x3, Ptr.Data);
   SpillStaticRegs();
   blr(x3);
   FillStaticRegs();
@@ -485,7 +491,6 @@ 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(GUESTRETURN,       GuestReturn);
   REGISTER_OP(SIGNALRETURN,      SignalReturn);
   REGISTER_OP(CALLBACKRETURN,    CallbackReturn);
   REGISTER_OP(EXITFUNCTION,      ExitFunction);

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

@@ -16,19 +16,19 @@ DEF_OP(VInsGPR) {
   mov(GetDst(Node), GetSrc(Op->Header.Args[0].ID()));
   switch (Op->Header.ElementSize) {
     case 1: {
-      ins(GetDst(Node).V16B(), Op->Index, GetReg<RA_32>(Op->Header.Args[1].ID()));
+      ins(GetDst(Node).V16B(), Op->DestIdx, GetReg<RA_32>(Op->Header.Args[1].ID()));
     break;
     }
     case 2: {
-      ins(GetDst(Node).V8H(), Op->Index, GetReg<RA_32>(Op->Header.Args[1].ID()));
+      ins(GetDst(Node).V8H(), Op->DestIdx, GetReg<RA_32>(Op->Header.Args[1].ID()));
     break;
     }
     case 4: {
-      ins(GetDst(Node).V4S(), Op->Index, GetReg<RA_32>(Op->Header.Args[1].ID()));
+      ins(GetDst(Node).V4S(), Op->DestIdx, GetReg<RA_32>(Op->Header.Args[1].ID()));
     break;
     }
     case 8: {
-      ins(GetDst(Node).V2D(), Op->Index, GetReg<RA_64>(Op->Header.Args[1].ID()));
+      ins(GetDst(Node).V2D(), Op->DestIdx, GetReg<RA_64>(Op->Header.Args[1].ID()));
     break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;

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

@@ -54,7 +54,7 @@ DEF_OP(AESDecLast) {
 DEF_OP(AESKeyGenAssist) {
 	auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();
 
-  aarch64::Label Constant;
+  aarch64::Literal ConstantLiteral (0x0C030609'0306090CULL, 0x040B0E01'0B0E0104ULL);
   aarch64::Label PastConstant;
 
   // Do a "regular" AESE step
@@ -63,8 +63,7 @@ DEF_OP(AESKeyGenAssist) {
   aese(VTMP1.V16B(), VTMP2.V16B());
 
   // Do a table shuffle to undo ShiftRows
-  adr(TMP1.X(), &Constant);
-  ldr(VTMP3, MemOperand(TMP1.X()));
+  ldr(VTMP3, &ConstantLiteral);
 
   // Now EOR in the RCON
   if (Op->RCON) {
@@ -80,14 +79,29 @@ DEF_OP(AESKeyGenAssist) {
   }
 
   b(&PastConstant);
-  bind(&Constant);
-  dc32(0x0B0E0104);
-  dc32(0x040B0E01);
-  dc32(0x0306090C);
-  dc32(0x0C030609);
+  place(&ConstantLiteral);
   bind(&PastConstant);
 }
 
+DEF_OP(CRC32) {
+  auto Op = IROp->C<IR::IROp_CRC32>();
+  switch (Op->SrcSize) {
+    case 1:
+      crc32cb(GetReg<RA_32>(Node), GetReg<RA_32>(Op->Src1.ID()), GetReg<RA_32>(Op->Src2.ID()));
+      break;
+    case 2:
+      crc32ch(GetReg<RA_32>(Node), GetReg<RA_32>(Op->Src1.ID()), GetReg<RA_32>(Op->Src2.ID()));
+      break;
+    case 4:
+      crc32cw(GetReg<RA_32>(Node), GetReg<RA_32>(Op->Src1.ID()), GetReg<RA_32>(Op->Src2.ID()));
+      break;
+    case 8:
+      crc32cx(GetReg<RA_32>(Node), GetReg<RA_32>(Op->Src1.ID()), GetReg<RA_64>(Op->Src2.ID()));
+      break;
+    default: LOGMAN_MSG_A_FMT("Unknown CRC32 size: {}", Op->SrcSize);
+  }
+}
+
 #undef DEF_OP
 void Arm64JITCore::RegisterEncryptionHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
@@ -97,7 +111,7 @@ void Arm64JITCore::RegisterEncryptionHandlers() {
   REGISTER_OP(VAESDEC,     AESDec);
   REGISTER_OP(VAESDECLAST, AESDecLast);
   REGISTER_OP(VAESKEYGENASSIST, AESKeyGenAssist);
-
+  REGISTER_OP(CRC32,             CRC32);
 #undef REGISTER_OP
 }
 }

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

@@ -21,10 +21,13 @@ $end_info$
 
 #include "Interface/IR/Passes/RegisterAllocationPass.h"
 
+#include "Utils/MemberFunctionToPointer.h"
+
 #include <FEXCore/Core/X86Enums.h>
 #include <FEXCore/Core/UContext.h>
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/CompilerDefs.h>
+
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 
 #include <sys/mman.h>
@@ -32,6 +35,40 @@ $end_info$
 #include <unistd.h>
 #include <string.h>
 
+namespace {
+static uint64_t LUDIV(uint64_t SrcHigh, uint64_t SrcLow, uint64_t Divisor) {
+  __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
+  __uint128_t Res = Source / Divisor;
+  return Res;
+}
+
+static int64_t LDIV(int64_t SrcHigh, int64_t SrcLow, int64_t Divisor) {
+  __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
+  __int128_t Res = Source / Divisor;
+  return Res;
+}
+
+static uint64_t LUREM(uint64_t SrcHigh, uint64_t SrcLow, uint64_t Divisor) {
+  __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
+  __uint128_t Res = Source % Divisor;
+  return Res;
+}
+
+static int64_t LREM(int64_t SrcHigh, int64_t SrcLow, int64_t Divisor) {
+  __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
+  __int128_t Res = Source % Divisor;
+  return Res;
+}
+
+static void PrintValue(uint64_t Value) {
+  LogMan::Msg::DFmt("Value: 0x{:x}", Value);
+}
+
+static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
+  LogMan::Msg::DFmt("Value: 0x{:016x}'{:016x}", ValueUpper, Value);
+}
+}
+
 namespace FEXCore::CPU {
 
 void Arm64JITCore::CopyNecessaryDataForCompileThread(CPUBackend *Original) {
@@ -56,8 +93,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushDynamicRegsAndLR();
 
         uxth(w0, GetReg<RA_32>(IROp->Args[0].ID()));
-        LoadConstant(x1, (uintptr_t)Info.fn);
-
+        ldr(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x1);
 
         PopDynamicRegsAndLR();
@@ -72,8 +108,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushDynamicRegsAndLR();
 
         fmov(v0.S(), GetSrc(IROp->Args[0].ID()).S()) ;
-        LoadConstant(x0, (uintptr_t)Info.fn);
-
+        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x0);
 
         PopDynamicRegsAndLR();
@@ -92,8 +127,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushDynamicRegsAndLR();
 
         mov(v0.D(), GetSrc(IROp->Args[0].ID()).D());
-        LoadConstant(x0, (uintptr_t)Info.fn);
-
+        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x0);
 
         PopDynamicRegsAndLR();
@@ -118,8 +152,7 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         else {
           mov(w0, GetReg<RA_32>(IROp->Args[0].ID()));
         }
-        LoadConstant(x1, (uintptr_t)Info.fn);
-
+        ldr(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x1);
 
         PopDynamicRegsAndLR();
@@ -140,8 +173,7 @@ 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);
 
-        LoadConstant(x2, (uintptr_t)Info.fn);
-
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -160,8 +192,7 @@ 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);
 
-        LoadConstant(x2, (uintptr_t)Info.fn);
-
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -180,8 +211,7 @@ 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);
 
-        LoadConstant(x2, (uintptr_t)Info.fn);
-
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -199,8 +229,7 @@ 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);
 
-        LoadConstant(x2, (uintptr_t)Info.fn);
-
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -218,8 +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);
 
-        LoadConstant(x2, (uintptr_t)Info.fn);
-
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -240,8 +268,7 @@ 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);
 
-        LoadConstant(x4, (uintptr_t)Info.fn);
-
+        ldr(x4, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x4);
 
         PopDynamicRegsAndLR();
@@ -259,8 +286,7 @@ 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);
 
-        LoadConstant(x2, (uintptr_t)Info.fn);
-
+        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x2);
 
         PopDynamicRegsAndLR();
@@ -283,8 +309,7 @@ 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);
 
-        LoadConstant(x4, (uintptr_t)Info.fn);
-
+        ldr(x4, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.FallbackHandlerPointers[Info.HandlerIndex])));
         blr(x4);
 
         PopDynamicRegsAndLR();
@@ -336,24 +361,6 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::Intern
   : Arm64Emitter(ctx, 0)
   , CTX {ctx}
   , ThreadState {Thread} {
-  {
-    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;
-
   RAPass = Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA");
 
 #if DEBUG
@@ -393,44 +400,100 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::Intern
   RegisterVectorHandlers();
   RegisterEncryptionHandlers();
 
+  {
+    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;
+  }
+
   if (!CompileThread) {
     ThreadSharedData.SignalHandlerRefCounterPtr = &Dispatcher->SignalHandlerRefCounter;
     ThreadSharedData.SignalReturnInstruction = 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 {
-      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);
-    }
   }
+
+  {
+    // 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);
+  }
+
+  // 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);
+  EmitDetectionString();
+
+  CurrentCodeBuffer = &InitialCodeBuffer;
+}
+
+void Arm64JITCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
+  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);
+  }
+}
+
+void Arm64JITCore::EmitDetectionString() {
+  const char JITString[] = "FEXJIT::Arm64JITCore::";
+  auto Buffer = GetBuffer();
+  Buffer->EmitString(JITString);
+  Buffer->Align();
 }
 
 void Arm64JITCore::ClearCache() {
@@ -473,6 +536,7 @@ void Arm64JITCore::ClearCache() {
     EmplaceNewCodeBuffer(NewCodeBuffer);
     *Buffer = vixl::CodeBuffer(NewCodeBuffer.Ptr, NewCodeBuffer.Size);
   }
+  EmitDetectionString();
 }
 
 Arm64JITCore::~Arm64JITCore() {
@@ -582,7 +646,12 @@ bool Arm64JITCore::IsInlineEntrypointOffset(const IR::OrderedNodeWrapper& WNode,
   if (OpHeader->Op == IR::IROps::OP_INLINEENTRYPOINTOFFSET) {
     auto Op = OpHeader->C<IR::IROp_InlineEntrypointOffset>();
     if (Value) {
-      *Value = Entry + Op->Offset;
+      uint64_t Mask = ~0ULL;
+      uint8_t OpSize = OpHeader->Size;
+      if (OpSize == 4) {
+        Mask = 0xFFFF'FFFFULL;
+      }
+      *Value = (Entry + Op->Offset) & Mask;
     }
     return true;
   } else {
@@ -668,7 +737,7 @@ void *Arm64JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IR
       str(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
 
       // Stop the thread
-      LoadConstant(x0, ThreadSharedData.Dispatcher->ThreadPauseHandlerAddressSpillSRA);
+      ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.ThreadPauseHandlerSpillSRA)));
       br(x0);
     }
     bind(&RunBlock);
@@ -696,6 +765,7 @@ 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
 
+    uintptr_t BlockStartHostCode = GetCursorAddress<uintptr_t>();
     {
       const auto Node = IR->GetID(BlockNode);
       const auto IsTarget = JumpTargets.try_emplace(Node).first;
@@ -710,10 +780,6 @@ 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);
 
@@ -723,7 +789,7 @@ 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({BlockStartHostCode, static_cast<uint32_t>(GetCursorAddress<uintptr_t>() - BlockStartHostCode)});
     }
   }
 
@@ -800,4 +866,8 @@ uint64_t Arm64JITCore::ExitFunctionLink(Arm64JITCore *core, FEXCore::Core::CpuSt
 std::unique_ptr<CPUBackend> CreateArm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread) {
   return std::make_unique<Arm64JITCore>(ctx, Thread, CompileThread);
 }
+
+void InitializeArm64JITSignalHandlers(FEXCore::Context::Context *CTX) {
+  Arm64JITCore::InitializeSignalHandlers(CTX);
+}
 }

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

@@ -68,6 +68,8 @@ public:
     return Dispatcher->IsAddressInJITCode(Address, IncludeDispatcher, IncludeCompileService);
   }
 
+  static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
+
 private:
   FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
 
@@ -183,6 +185,9 @@ private:
   };
 
   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;
 
@@ -277,7 +282,6 @@ private:
   ///< Branch ops
   DEF_OP(GuestCallDirect);
   DEF_OP(GuestCallIndirect);
-  DEF_OP(GuestReturn);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -305,7 +309,7 @@ private:
   DEF_OP(GetHostFlag);
 
   ///< Memory ops
-    DEF_OP(LoadContext);
+  DEF_OP(LoadContext);
   DEF_OP(StoreContext);
   DEF_OP(LoadRegister);
   DEF_OP(StoreRegister);
@@ -336,6 +340,7 @@ private:
   DEF_OP(GetRoundingMode);
   DEF_OP(SetRoundingMode);
   DEF_OP(ProcessorID);
+  DEF_OP(RDRAND);
 
   ///< Move ops
   DEF_OP(ExtractElementPair);
@@ -345,8 +350,6 @@ private:
   ///< Vector ops
   DEF_OP(VectorZero);
   DEF_OP(VectorImm);
-  DEF_OP(CreateVector2);
-  DEF_OP(CreateVector4);
   DEF_OP(SplatVector2);
   DEF_OP(SplatVector4);
   DEF_OP(VMov);
@@ -434,6 +437,7 @@ private:
   DEF_OP(VSMull2);
   DEF_OP(VUABDL);
   DEF_OP(VTBL1);
+  DEF_OP(VRev64);
 
   ///< Encryption ops
   DEF_OP(AESImc);
@@ -442,6 +446,7 @@ private:
   DEF_OP(AESDec);
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
+  DEF_OP(CRC32);
 #undef DEF_OP
 };
 

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

@@ -698,29 +698,29 @@ DEF_OP(LoadMemTSO) {
 DEF_OP(StoreMem) {
   auto Op = IROp->C<IR::IROp_StoreMem>();
 
-  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) {
     switch (IROp->Size) {
       case 1:
-        strb(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+        strb(GetReg<RA_64>(Op->Value.ID()), MemSrc);
         break;
       case 2:
-        strh(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+        strh(GetReg<RA_64>(Op->Value.ID()), MemSrc);
         break;
       case 4:
-        str(GetReg<RA_32>(Op->Header.Args[1].ID()), MemSrc);
+        str(GetReg<RA_32>(Op->Value.ID()), MemSrc);
         break;
       case 8:
-        str(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+        str(GetReg<RA_64>(Op->Value.ID()), MemSrc);
         break;
       default:  LOGMAN_MSG_A_FMT("Unhandled StoreMem size: {}", IROp->Size);
     }
   }
   else {
-    auto Src = GetSrc(Op->Header.Args[1].ID());
+    auto Src = GetSrc(Op->Value.ID());
     switch (IROp->Size) {
       case 1:
         str(Src.B(), MemSrc);
@@ -744,7 +744,7 @@ DEF_OP(StoreMem) {
 
 DEF_OP(StoreMemTSO) {
   auto Op = IROp->C<IR::IROp_StoreMemTSO>();
-  auto MemSrc = MemOperand(GetReg<RA_64>(Op->Header.Args[0].ID()));
+  auto MemSrc = MemOperand(GetReg<RA_64>(Op->Addr.ID()));
 
   if (!Op->Offset.IsInvalid()) {
     LOGMAN_MSG_A_FMT("StoreMemTSO: No offset allowed");
@@ -753,19 +753,19 @@ DEF_OP(StoreMemTSO) {
   if (Op->Class == FEXCore::IR::GPRClass) {
     if (IROp->Size == 1) {
       // 8bit load is always aligned to natural alignment
-      stlrb(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+      stlrb(GetReg<RA_64>(Op->Value.ID()), MemSrc);
     }
     else {
       nop();
       switch (IROp->Size) {
         case 2:
-          stlrh(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+          stlrh(GetReg<RA_64>(Op->Value.ID()), MemSrc);
           break;
         case 4:
-          stlr(GetReg<RA_32>(Op->Header.Args[1].ID()), MemSrc);
+          stlr(GetReg<RA_32>(Op->Value.ID()), MemSrc);
           break;
         case 8:
-          stlr(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+          stlr(GetReg<RA_64>(Op->Value.ID()), MemSrc);
           break;
         default:  LOGMAN_MSG_A_FMT("Unhandled StoreMemTSO size: {}", IROp->Size);
       }
@@ -774,7 +774,7 @@ DEF_OP(StoreMemTSO) {
   }
   else {
     dmb(InnerShareable, BarrierAll);
-    auto Src = GetSrc(Op->Header.Args[1].ID());
+    auto Src = GetSrc(Op->Value.ID());
     switch (IROp->Size) {
       case 1:
         str(Src.B(), MemSrc);
@@ -800,7 +800,7 @@ DEF_OP(StoreMemTSO) {
 DEF_OP(ParanoidLoadMemTSO) {
   auto Op = IROp->C<IR::IROp_LoadMemTSO>();
 
-  auto MemSrc = MemOperand(GetReg<RA_64>(Op->Header.Args[0].ID()));
+  auto MemSrc = MemOperand(GetReg<RA_64>(Op->Addr.ID()));
 
   if (!Op->Offset.IsInvalid()) {
     LOGMAN_MSG_A_FMT("ParanoidLoadMemTSO: No offset allowed");
@@ -857,7 +857,7 @@ DEF_OP(ParanoidLoadMemTSO) {
 
 DEF_OP(ParanoidStoreMemTSO) {
   auto Op = IROp->C<IR::IROp_StoreMemTSO>();
-  auto MemSrc = MemOperand(GetReg<RA_64>(Op->Header.Args[0].ID()));
+  auto MemSrc = MemOperand(GetReg<RA_64>(Op->Addr.ID()));
 
   if (!Op->Offset.IsInvalid()) {
     LOGMAN_MSG_A_FMT("ParanoidStoreMemTSO: No offset allowed");
@@ -866,25 +866,25 @@ DEF_OP(ParanoidStoreMemTSO) {
   if (Op->Class == FEXCore::IR::GPRClass) {
     if (IROp->Size == 1) {
       // 8bit load is always aligned to natural alignment
-      stlrb(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+      stlrb(GetReg<RA_64>(Op->Value.ID()), MemSrc);
     }
     else {
       switch (IROp->Size) {
         case 2:
-          stlrh(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+          stlrh(GetReg<RA_64>(Op->Value.ID()), MemSrc);
           break;
         case 4:
-          stlr(GetReg<RA_32>(Op->Header.Args[1].ID()), MemSrc);
+          stlr(GetReg<RA_32>(Op->Value.ID()), MemSrc);
           break;
         case 8:
-          stlr(GetReg<RA_64>(Op->Header.Args[1].ID()), MemSrc);
+          stlr(GetReg<RA_64>(Op->Value.ID()), MemSrc);
           break;
         default:  LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", IROp->Size);
       }
     }
   }
   else {
-    auto Src = GetSrc(Op->Header.Args[1].ID());
+    auto Src = GetSrc(Op->Value.ID());
     if (IROp->Size == 1) {
       // 8bit load is always aligned to natural alignment
       mov(TMP1.W(), Src.V16B(), 0);

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

@@ -7,14 +7,6 @@ $end_info$
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 
 namespace FEXCore::CPU {
-static void PrintValue(uint64_t Value) {
-  LogMan::Msg::DFmt("Value: 0x{:x}", Value);
-}
-
-static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
-  LogMan::Msg::DFmt("Value: 0x{:016x}'{:016x}", ValueUpper, Value);
-}
-
 using namespace vixl;
 using namespace vixl::aarch64;
 #define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
@@ -44,7 +36,7 @@ DEF_OP(Break) {
       break;
     case FEXCore::IR::Break_Overflow: // overflow
       ResetStack();
-      LoadConstant(TMP1, ThreadSharedData.Dispatcher->OverflowExceptionInstructionAddress);
+      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.OverflowExceptionHandler)));
       br(TMP1);
       break;
     case FEXCore::IR::Break_Halt: { // HLT
@@ -54,14 +46,13 @@ DEF_OP(Break) {
       add(sp, TMP1, 0);
 
       // Now we need to jump to the thread stop handler
-      LoadConstant(TMP1, ThreadSharedData.Dispatcher->ThreadStopHandlerAddressSpillSRA);
+      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.ThreadStopHandlerSpillSRA)));
       br(TMP1);
       break;
     }
     case FEXCore::IR::Break_Interrupt3: { // INT3
       ResetStack();
-
-      LoadConstant(TMP1, ThreadSharedData.Dispatcher->ThreadPauseHandlerAddressSpillSRA);
+      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.ThreadPauseHandlerSpillSRA)));
       br(TMP1);
       break;
     }
@@ -69,7 +60,7 @@ DEF_OP(Break) {
     {
       ResetStack();
 
-      LoadConstant(TMP1, ThreadSharedData.Dispatcher->UnimplementedInstructionAddress);
+      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.UnimplementedInstructionHandler)));
       br(TMP1);
 
       break;
@@ -143,13 +134,13 @@ DEF_OP(Print) {
 
   if (IsGPR(Op->Header.Args[0].ID())) {
     mov(x0, GetReg<RA_64>(Op->Header.Args[0].ID()));
-    LoadConstant(x3, reinterpret_cast<uint64_t>(PrintValue));
+    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.PrintValue)));
   }
   else {
     fmov(x0, GetSrc(Op->Header.Args[0].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);
-    LoadConstant(x3, reinterpret_cast<uint64_t>(PrintVectorValue));
+    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.PrintVectorValue)));
   }
   SpillStaticRegs();
   blr(x3);
@@ -210,6 +201,24 @@ DEF_OP(ProcessorID) {
   orr(GetReg<RA_64>(Node), x0, Operand(x1, LSL, 12));
 }
 
+DEF_OP(RDRAND) {
+  auto Op = IROp->C<IR::IROp_RDRAND>();
+
+  // Results are in x0, x1
+  // Results want to be in a i64v2 vector
+  auto Dst = GetSrcPair<RA_64>(Node);
+
+  if (Op->GetReseeded) {
+    mrs(Dst.first, RNDRRS);
+  }
+  else {
+    mrs(Dst.first, RNDR);
+  }
+
+  // If the rng number is valid then NZCV is 0b0000, otherwise NZCV is 0b0100
+  cset(Dst.second, Condition::ne);
+}
+
 #undef DEF_OP
 void Arm64JITCore::RegisterMiscHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
@@ -227,6 +236,8 @@ void Arm64JITCore::RegisterMiscHandlers() {
   REGISTER_OP(SETROUNDINGMODE, SetRoundingMode);
   REGISTER_OP(INVALIDATEFLAGS,   NoOp);
   REGISTER_OP(PROCESSORID,   ProcessorID);
+  REGISTER_OP(RDRAND, RDRAND);
+
 #undef REGISTER_OP
 }
 }

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

@@ -37,7 +37,7 @@ DEF_OP(CreateElementPair) {
   aarch64::Register RegSecond;
   aarch64::Register RegTmp;
 
-  switch (Op->Header.Size) {
+  switch (IROp->ElementSize) {
     case 4: {
       Dst = GetSrcPair<RA_32>(Node);
       RegFirst = GetReg<RA_32>(Op->Header.Args[0].ID());

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

@@ -42,14 +42,6 @@ DEF_OP(VectorImm) {
   }
 }
 
-DEF_OP(CreateVector2) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
-DEF_OP(CreateVector4) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
 DEF_OP(SplatVector2) {
   auto Op = IROp->C<IR::IROp_SplatVector2>();
   uint8_t OpSize = IROp->Size;
@@ -1758,8 +1750,7 @@ DEF_OP(VInsScalarElement) {
 
 DEF_OP(VExtractElement) {
   auto Op = IROp->C<IR::IROp_VExtractElement>();
-  uint8_t OpSize = IROp->Size;
-  switch (OpSize) {
+  switch (Op->Header.Size) {
     case 1:
       mov(GetDst(Node).B(), GetSrc(Op->Header.Args[0].ID()).V16B(), Op->Index);
     break;
@@ -1772,7 +1763,7 @@ DEF_OP(VExtractElement) {
     case 8:
       mov(GetDst(Node).D(), GetSrc(Op->Header.Args[0].ID()).V2D(), Op->Index);
     break;
-    default:  LOGMAN_MSG_A_FMT("Unhandled VExtractElement element size: {}", OpSize);
+    default:  LOGMAN_MSG_A_FMT("Unhandled VExtractElement element size: {}", Op->Header.Size);
   }
 }
 
@@ -2318,13 +2309,27 @@ DEF_OP(VTBL1) {
   }
 }
 
+DEF_OP(VRev64) {
+  auto Op = IROp->C<IR::IROp_VRev64>();
+  uint8_t OpSize = IROp->Size;
+  uint8_t Elements = OpSize / Op->Header.ElementSize;
+  // Vector
+  switch (Op->Header.ElementSize) {
+    case 1:
+    case 2:
+    case 4:
+      rev64(GetDst(Node).VCast(OpSize * 8, Elements), GetSrc(Op->Header.Args[0].ID()).VCast(OpSize * 8, Elements));
+      break;
+    case 8:
+    default: LOGMAN_MSG_A_FMT("Invalid Element Size: {}", Op->Header.ElementSize); break;
+  }
+}
+
 #undef DEF_OP
 void Arm64JITCore::RegisterVectorHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
   REGISTER_OP(VECTORZERO,        VectorZero);
   REGISTER_OP(VECTORIMM,         VectorImm);
-  REGISTER_OP(CREATEVECTOR2,     CreateVector2);
-  REGISTER_OP(CREATEVECTOR4,     CreateVector4);
   REGISTER_OP(SPLATVECTOR2,      SplatVector2);
   REGISTER_OP(SPLATVECTOR4,      SplatVector4);
   REGISTER_OP(VMOV,              VMov);
@@ -2413,6 +2418,7 @@ void Arm64JITCore::RegisterVectorHandlers() {
   REGISTER_OP(VSMULL2,           VSMull2);
   REGISTER_OP(VUABDL,            VUABDL);
   REGISTER_OP(VTBL1,             VTBL1);
+  REGISTER_OP(VREV64,            VRev64);
 #undef REGISTER_OP
 }
 }

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

@@ -16,8 +16,11 @@ class CPUBackend;
 [[nodiscard]] std::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::Context *ctx,
                                                            FEXCore::Core::InternalThreadState *Thread,
                                                            bool CompileThread);
+void InitializeX86JITSignalHandlers(FEXCore::Context::Context *CTX);
 
 [[nodiscard]] std::unique_ptr<CPUBackend> CreateArm64JITCore(FEXCore::Context::Context *ctx,
                                                              FEXCore::Core::InternalThreadState *Thread,
                                                              bool CompileThread);
+void InitializeArm64JITSignalHandlers(FEXCore::Context::Context *CTX);
+
 } // namespace FEXCore::CPU

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

@@ -24,7 +24,7 @@ namespace FEXCore::CPU {
 DEF_OP(TruncElementPair) {
   auto Op = IROp->C<IR::IROp_TruncElementPair>();
 
-  switch (Op->Size) {
+  switch (IROp->Size) {
     case 4: {
       auto Dst = GetSrcPair<RA_32>(Node);
       auto Src = GetSrcPair<RA_32>(Op->Header.Args[0].ID());
@@ -32,7 +32,7 @@ DEF_OP(TruncElementPair) {
       mov(Dst.second, Src.second);
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unhandled Truncation size: {}", Op->Size); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled Truncation size: {}", IROp->Size); break;
   }
 }
 
@@ -45,7 +45,13 @@ DEF_OP(EntrypointOffset) {
   auto Op = IROp->C<IR::IROp_EntrypointOffset>();
 
   auto Constant = Entry + Op->Offset;
-  mov(GetDst<RA_64>(Node), Constant);
+  uint64_t Mask = ~0ULL;
+  uint8_t OpSize = IROp->Size;
+  if (OpSize == 4) {
+    Mask = 0xFFFF'FFFFULL;
+  }
+
+  mov(GetDst<RA_64>(Node), Constant & Mask);
 }
 
 DEF_OP(InlineConstant) {
@@ -1144,19 +1150,19 @@ DEF_OP(VExtractToGPR) {
 
   switch (Op->Header.ElementSize) {
     case 1: {
-      pextrb(GetDst<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()), Op->Idx);
+      pextrb(GetDst<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()), Op->Index);
     break;
     }
     case 2: {
-      pextrw(GetDst<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()), Op->Idx);
+      pextrw(GetDst<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()), Op->Index);
     break;
     }
     case 4: {
-      pextrd(GetDst<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()), Op->Idx);
+      pextrd(GetDst<RA_32>(Node), GetSrc(Op->Header.Args[0].ID()), Op->Index);
     break;
     }
     case 8: {
-      pextrq(GetDst<RA_64>(Node), GetSrc(Op->Header.Args[0].ID()), Op->Idx);
+      pextrq(GetDst<RA_64>(Node), GetSrc(Op->Header.Args[0].ID()), Op->Index);
     break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;

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

@@ -18,20 +18,16 @@ namespace FEXCore::CPU {
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 DEF_OP(CASPair) {
   auto Op = IROp->C<IR::IROp_CAS>();
-  uint8_t OpSize = IROp->Size;
 
-  // Args[0]: Desired
-  // Args[1]: Expected
-  // Args[2]: Pointer
   // DataSrc = *Src1
   // if (DataSrc == Src3) { *Src1 == Src2; } Src2 = DataSrc
   // This will write to memory! Careful!
   // Third operand must be a calculated guest memory address
   //OrderedNode *CASResult = _CAS(Src3, Src2, Src1);
   auto Dst = GetSrcPair<RA_64>(Node);
-  auto Expected = GetSrcPair<RA_64>(Op->Header.Args[0].ID());
-  auto Desired = GetSrcPair<RA_64>(Op->Header.Args[1].ID());
-  auto MemSrc = GetSrc<RA_64>(Op->Header.Args[2].ID());
+  auto Expected = GetSrcPair<RA_64>(Op->Expected.ID());
+  auto Desired = GetSrcPair<RA_64>(Op->Desired.ID());
+  auto MemSrc = GetSrc<RA_64>(Op->Addr.ID());
 
   Xbyak::Reg MemReg = MemSrc;
 
@@ -47,7 +43,7 @@ DEF_OP(CASPair) {
 
   lock();
 
-  switch (OpSize) {
+  switch (IROp->ElementSize) {
     case 4: {
       cmpxchg8b(dword [MemReg]);
       // EDX:EAX now contains the result
@@ -62,7 +58,7 @@ DEF_OP(CASPair) {
       mov(Dst.second, rdx);
     break;
     }
-    default: LOGMAN_MSG_A_FMT("Unsupported: {}", OpSize);
+    default: LOGMAN_MSG_A_FMT("Unsupported: {}", IROp->ElementSize);
   }
 }
 
@@ -70,18 +66,15 @@ DEF_OP(CAS) {
   auto Op = IROp->C<IR::IROp_CAS>();
   uint8_t OpSize = IROp->Size;
 
-  // Args[0]: Desired
-  // Args[1]: Expected
-  // Args[2]: Pointer
   // DataSrc = *Src1
   // if (DataSrc == Src3) { *Src1 == Src2; } Src2 = DataSrc
   // This will write to memory! Careful!
   // Third operand must be a calculated guest memory address
   //OrderedNode *CASResult = _CAS(Src3, Src2, Src1);
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[2].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
 
-  mov(rax, GetSrc<RA_64>(Op->Header.Args[0].ID()));
+  mov(rax, GetSrc<RA_64>(Op->Expected.ID()));
 
   // RCX now contains pointer
   // RAX contains our expected value
@@ -90,23 +83,23 @@ DEF_OP(CAS) {
   lock();
   switch (OpSize) {
   case 1: {
-    cmpxchg(byte [MemReg], GetSrc<RA_8>(Op->Header.Args[1].ID()));
+    cmpxchg(byte [MemReg], GetSrc<RA_8>(Op->Desired.ID()));
     movzx(GetDst<RA_64>(Node), al);
   break;
   }
   case 2: {
-    cmpxchg(word [MemReg], GetSrc<RA_16>(Op->Header.Args[1].ID()));
+    cmpxchg(word [MemReg], GetSrc<RA_16>(Op->Desired.ID()));
     movzx(GetDst<RA_64>(Node), ax);
   break;
   }
   case 4: {
-    cmpxchg(dword [MemReg], GetSrc<RA_32>(Op->Header.Args[1].ID()));
+    cmpxchg(dword [MemReg], GetSrc<RA_32>(Op->Desired.ID()));
     // RAX now contains the result
     mov (GetDst<RA_64>(Node), eax);
   break;
   }
   case 8: {
-    cmpxchg(qword [MemReg], GetSrc<RA_64>(Op->Header.Args[1].ID()));
+    cmpxchg(qword [MemReg], GetSrc<RA_64>(Op->Desired.ID()));
     // RAX now contains the result
     mov (GetDst<RA_64>(Node), rax);
   break;
@@ -118,21 +111,21 @@ DEF_OP(CAS) {
 DEF_OP(AtomicAdd) {
   auto Op = IROp->C<IR::IROp_AtomicAdd>();
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
 
   lock();
   switch (IROp->Size) {
     case 1:
-      add(byte [MemReg], GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      add(byte [MemReg], GetSrc<RA_8>(Op->Value.ID()));
       break;
     case 2:
-      add(word [MemReg], GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      add(word [MemReg], GetSrc<RA_16>(Op->Value.ID()));
       break;
     case 4:
-      add(dword [MemReg], GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      add(dword [MemReg], GetSrc<RA_32>(Op->Value.ID()));
       break;
     case 8:
-      add(qword [MemReg], GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      add(qword [MemReg], GetSrc<RA_64>(Op->Value.ID()));
       break;
     default:  LOGMAN_MSG_A_FMT("Unhandled AtomicAdd size: {}", IROp->Size);
   }
@@ -141,20 +134,20 @@ DEF_OP(AtomicAdd) {
 DEF_OP(AtomicSub) {
   auto Op = IROp->C<IR::IROp_AtomicSub>();
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   lock();
   switch (IROp->Size) {
     case 1:
-      sub(byte [MemReg], GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      sub(byte [MemReg], GetSrc<RA_8>(Op->Value.ID()));
       break;
     case 2:
-      sub(word [MemReg], GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      sub(word [MemReg], GetSrc<RA_16>(Op->Value.ID()));
       break;
     case 4:
-      sub(dword [MemReg], GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      sub(dword [MemReg], GetSrc<RA_32>(Op->Value.ID()));
       break;
     case 8:
-      sub(qword [MemReg], GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      sub(qword [MemReg], GetSrc<RA_64>(Op->Value.ID()));
       break;
     default:  LOGMAN_MSG_A_FMT("Unhandled AtomicAdd size: {}", IROp->Size);
   }
@@ -163,20 +156,20 @@ DEF_OP(AtomicSub) {
 DEF_OP(AtomicAnd) {
   auto Op = IROp->C<IR::IROp_AtomicAnd>();
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   lock();
   switch (IROp->Size) {
     case 1:
-      and_(byte [MemReg], GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      and_(byte [MemReg], GetSrc<RA_8>(Op->Value.ID()));
       break;
     case 2:
-      and_(word [MemReg], GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      and_(word [MemReg], GetSrc<RA_16>(Op->Value.ID()));
       break;
     case 4:
-      and_(dword [MemReg], GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      and_(dword [MemReg], GetSrc<RA_32>(Op->Value.ID()));
       break;
     case 8:
-      and_(qword [MemReg], GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      and_(qword [MemReg], GetSrc<RA_64>(Op->Value.ID()));
       break;
     default:  LOGMAN_MSG_A_FMT("Unhandled AtomicAdd size: {}", IROp->Size);
   }
@@ -185,20 +178,20 @@ DEF_OP(AtomicAnd) {
 DEF_OP(AtomicOr) {
   auto Op = IROp->C<IR::IROp_AtomicOr>();
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   lock();
   switch (IROp->Size) {
     case 1:
-      or_(byte [MemReg], GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      or_(byte [MemReg], GetSrc<RA_8>(Op->Value.ID()));
       break;
     case 2:
-      or_(word [MemReg], GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      or_(word [MemReg], GetSrc<RA_16>(Op->Value.ID()));
       break;
     case 4:
-      or_(dword [MemReg], GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      or_(dword [MemReg], GetSrc<RA_32>(Op->Value.ID()));
       break;
     case 8:
-      or_(qword [MemReg], GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      or_(qword [MemReg], GetSrc<RA_64>(Op->Value.ID()));
       break;
     default:  LOGMAN_MSG_A_FMT("Unhandled AtomicAdd size: {}", IROp->Size);
   }
@@ -207,20 +200,20 @@ DEF_OP(AtomicOr) {
 DEF_OP(AtomicXor) {
   auto Op = IROp->C<IR::IROp_AtomicXor>();
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   lock();
   switch (IROp->Size) {
     case 1:
-      xor_(byte [MemReg], GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      xor_(byte [MemReg], GetSrc<RA_8>(Op->Value.ID()));
       break;
     case 2:
-      xor_(word [MemReg], GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      xor_(word [MemReg], GetSrc<RA_16>(Op->Value.ID()));
       break;
     case 4:
-      xor_(dword [MemReg], GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      xor_(dword [MemReg], GetSrc<RA_32>(Op->Value.ID()));
       break;
     case 8:
-      xor_(qword [MemReg], GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      xor_(qword [MemReg], GetSrc<RA_64>(Op->Value.ID()));
       break;
     default:  LOGMAN_MSG_A_FMT("Unhandled AtomicAdd size: {}", IROp->Size);
   }
@@ -230,26 +223,26 @@ DEF_OP(AtomicSwap) {
   auto Op = IROp->C<IR::IROp_AtomicSwap>();
 
   Xbyak::Reg MemReg = rax;
-  mov(MemReg, GetSrc<RA_64>(Op->Header.Args[0].ID()));
+  mov(MemReg, GetSrc<RA_64>(Op->Addr.ID()));
 
   switch (IROp->Size) {
     case 1:
-      movzx(GetDst<RA_64>(Node), GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      movzx(GetDst<RA_64>(Node), GetSrc<RA_8>(Op->Value.ID()));
       lock();
       xchg(byte [MemReg], GetDst<RA_8>(Node));
       break;
     case 2:
-      movzx(GetDst<RA_64>(Node), GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      movzx(GetDst<RA_64>(Node), GetSrc<RA_16>(Op->Value.ID()));
       lock();
       xchg(word [MemReg], GetDst<RA_16>(Node));
       break;
     case 4:
-      mov(GetDst<RA_64>(Node), GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      mov(GetDst<RA_64>(Node), GetSrc<RA_32>(Op->Value.ID()));
       lock();
       xchg(dword [MemReg], GetDst<RA_32>(Node));
       break;
     case 8:
-      mov(GetDst<RA_64>(Node), GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      mov(GetDst<RA_64>(Node), GetSrc<RA_64>(Op->Value.ID()));
       lock();
       xchg(qword [MemReg], GetDst<RA_64>(Node));
       break;
@@ -260,28 +253,28 @@ DEF_OP(AtomicSwap) {
 DEF_OP(AtomicFetchAdd) {
   auto Op = IROp->C<IR::IROp_AtomicFetchAdd>();
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   switch (IROp->Size) {
     case 1:
-      movzx(rcx, GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      movzx(rcx, GetSrc<RA_8>(Op->Value.ID()));
       lock();
       xadd(byte [MemReg], cl);
       movzx(GetDst<RA_32>(Node), cl);
       break;
     case 2:
-      movzx(rcx, GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      movzx(rcx, GetSrc<RA_16>(Op->Value.ID()));
       lock();
       xadd(word [MemReg], cx);
       movzx(GetDst<RA_32>(Node), cx);
       break;
     case 4:
-      mov(ecx, GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      mov(ecx, GetSrc<RA_32>(Op->Value.ID()));
       lock();
       xadd(dword [MemReg], ecx);
       mov(GetDst<RA_64>(Node), ecx);
       break;
     case 8:
-      mov(rcx, GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      mov(rcx, GetSrc<RA_64>(Op->Value.ID()));
       lock();
       xadd(qword [MemReg], rcx);
       mov(GetDst<RA_64>(Node), rcx);
@@ -293,31 +286,31 @@ DEF_OP(AtomicFetchAdd) {
 DEF_OP(AtomicFetchSub) {
   auto Op = IROp->C<IR::IROp_AtomicFetchSub>();
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   switch (IROp->Size) {
     case 1:
-      mov(cl, GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      mov(cl, GetSrc<RA_8>(Op->Value.ID()));
       neg(cl);
       lock();
       xadd(byte [MemReg], cl);
       movzx(GetDst<RA_32>(Node), cl);
       break;
     case 2:
-      mov(cx, GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      mov(cx, GetSrc<RA_16>(Op->Value.ID()));
       neg(cx);
       lock();
       xadd(word [MemReg], cx);
       movzx(GetDst<RA_32>(Node), cx);
       break;
     case 4:
-      mov(ecx, GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      mov(ecx, GetSrc<RA_32>(Op->Value.ID()));
       neg(ecx);
       lock();
       xadd(dword [MemReg], ecx);
       mov(GetDst<RA_32>(Node), ecx);
       break;
     case 8:
-      mov(rcx, GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      mov(rcx, GetSrc<RA_64>(Op->Value.ID()));
       neg(rcx);
       lock();
       xadd(qword [MemReg], rcx);
@@ -331,7 +324,7 @@ DEF_OP(AtomicFetchAnd) {
   auto Op = IROp->C<IR::IROp_AtomicFetchAnd>();
 
   // TMP1 = rax
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
 
   switch (IROp->Size) {
     case 1: {
@@ -341,7 +334,7 @@ DEF_OP(AtomicFetchAnd) {
       L(Loop);
       mov(TMP2.cvt8(), TMP1.cvt8());
       mov(TMP3.cvt8(), TMP1.cvt8());
-      and_(TMP2.cvt8(), GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      and_(TMP2.cvt8(), GetSrc<RA_8>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(byte [MemReg], TMP2.cvt8());
@@ -357,7 +350,7 @@ DEF_OP(AtomicFetchAnd) {
       L(Loop);
       mov(TMP2.cvt16(), TMP1.cvt16());
       mov(TMP3.cvt16(), TMP1.cvt16());
-      and_(TMP2.cvt16(), GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      and_(TMP2.cvt16(), GetSrc<RA_16>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(word [MemReg], TMP2.cvt16());
@@ -374,7 +367,7 @@ DEF_OP(AtomicFetchAnd) {
       L(Loop);
       mov(TMP2.cvt32(), TMP1.cvt32());
       mov(TMP3.cvt32(), TMP1.cvt32());
-      and_(TMP2.cvt32(), GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      and_(TMP2.cvt32(), GetSrc<RA_32>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(dword [MemReg], TMP2.cvt32());
@@ -391,7 +384,7 @@ DEF_OP(AtomicFetchAnd) {
       L(Loop);
       mov(TMP2.cvt64(), TMP1.cvt64());
       mov(TMP3.cvt64(), TMP1.cvt64());
-      and_(TMP2.cvt64(), GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      and_(TMP2.cvt64(), GetSrc<RA_64>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(qword [MemReg], TMP2.cvt64());
@@ -409,7 +402,7 @@ DEF_OP(AtomicFetchOr) {
   auto Op = IROp->C<IR::IROp_AtomicFetchOr>();
 
   // TMP1 = rax
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   switch (IROp->Size) {
     case 1: {
       mov(TMP1.cvt8(), byte [MemReg]);
@@ -418,7 +411,7 @@ DEF_OP(AtomicFetchOr) {
       L(Loop);
       mov(TMP2.cvt8(), TMP1.cvt8());
       mov(TMP3.cvt8(), TMP1.cvt8());
-      or_(TMP2.cvt8(), GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      or_(TMP2.cvt8(), GetSrc<RA_8>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(byte [MemReg], TMP2.cvt8());
@@ -434,7 +427,7 @@ DEF_OP(AtomicFetchOr) {
       L(Loop);
       mov(TMP2.cvt16(), TMP1.cvt16());
       mov(TMP3.cvt16(), TMP1.cvt16());
-      or_(TMP2.cvt16(), GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      or_(TMP2.cvt16(), GetSrc<RA_16>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(word [MemReg], TMP2.cvt16());
@@ -451,7 +444,7 @@ DEF_OP(AtomicFetchOr) {
       L(Loop);
       mov(TMP2.cvt32(), TMP1.cvt32());
       mov(TMP3.cvt32(), TMP1.cvt32());
-      or_(TMP2.cvt32(), GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      or_(TMP2.cvt32(), GetSrc<RA_32>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(dword [MemReg], TMP2.cvt32());
@@ -468,7 +461,7 @@ DEF_OP(AtomicFetchOr) {
       L(Loop);
       mov(TMP2.cvt64(), TMP1.cvt64());
       mov(TMP3.cvt64(), TMP1.cvt64());
-      or_(TMP2.cvt64(), GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      or_(TMP2.cvt64(), GetSrc<RA_64>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(qword [MemReg], TMP2.cvt64());
@@ -486,7 +479,7 @@ DEF_OP(AtomicFetchXor) {
   auto Op = IROp->C<IR::IROp_AtomicFetchXor>();
 
   // TMP1 = rax
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   switch (IROp->Size) {
     case 1: {
       mov(TMP1.cvt8(), byte [MemReg]);
@@ -495,7 +488,7 @@ DEF_OP(AtomicFetchXor) {
       L(Loop);
       mov(TMP2.cvt8(), TMP1.cvt8());
       mov(TMP3.cvt8(), TMP1.cvt8());
-      xor_(TMP2.cvt8(), GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      xor_(TMP2.cvt8(), GetSrc<RA_8>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(byte [MemReg], TMP2.cvt8());
@@ -511,7 +504,7 @@ DEF_OP(AtomicFetchXor) {
       L(Loop);
       mov(TMP2.cvt16(), TMP1.cvt16());
       mov(TMP3.cvt16(), TMP1.cvt16());
-      xor_(TMP2.cvt16(), GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      xor_(TMP2.cvt16(), GetSrc<RA_16>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(word [MemReg], TMP2.cvt16());
@@ -528,7 +521,7 @@ DEF_OP(AtomicFetchXor) {
       L(Loop);
       mov(TMP2.cvt32(), TMP1.cvt32());
       mov(TMP3.cvt32(), TMP1.cvt32());
-      xor_(TMP2.cvt32(), GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      xor_(TMP2.cvt32(), GetSrc<RA_32>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(dword [MemReg], TMP2.cvt32());
@@ -545,7 +538,7 @@ DEF_OP(AtomicFetchXor) {
       L(Loop);
       mov(TMP2.cvt64(), TMP1.cvt64());
       mov(TMP3.cvt64(), TMP1.cvt64());
-      xor_(TMP2.cvt64(), GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      xor_(TMP2.cvt64(), GetSrc<RA_64>(Op->Value.ID()));
 
       // Updates RAX with the value from memory
       lock(); cmpxchg(qword [MemReg], TMP2.cvt64());
@@ -562,7 +555,7 @@ DEF_OP(AtomicFetchXor) {
 DEF_OP(AtomicFetchNeg) {
   auto Op = IROp->C<IR::IROp_AtomicFetchNeg>();
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
   switch (IROp->Size) {
     case 1: {
       mov(TMP1.cvt8(), byte [MemReg]);

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

@@ -37,18 +37,13 @@ DEF_OP(GuestCallIndirect) {
   LogMan::Msg::DFmt("Unimplemented");
 }
 
-DEF_OP(GuestReturn) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
 DEF_OP(SignalReturn) {
   // Adjust the stack first for a regular return
   if (SpillSlots) {
     add(rsp, SpillSlots * 16); // + 8 to consume return address
   }
 
-  mov(TMP1, ThreadSharedData.SignalHandlerReturnAddress);
-  jmp(TMP1);
+  jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalReturnHandler)]);
 }
 
 DEF_OP(CallbackReturn) {
@@ -58,8 +53,7 @@ DEF_OP(CallbackReturn) {
   }
 
   // Make sure to adjust the refcounter so we don't clear the cache now
-  mov(rax, reinterpret_cast<uint64_t>(ThreadSharedData.SignalHandlerRefCounterPtr));
-  sub(dword [rax], 1);
+  sub(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalHandlerRefCountPointer)], 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);
@@ -104,7 +98,8 @@ DEF_OP(ExitFunction) {
     Xbyak::Reg RipReg = GetSrc<RA_64>(Op->NewRIP.ID());
 
     // L1 Cache
-    mov(rcx, ThreadState->LookupCache->GetL1Pointer());
+    mov(rcx, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
+
     mov(rax, RipReg);
 
     and_(rax, LookupCache::L1_ENTRIES_MASK);
@@ -117,9 +112,8 @@ DEF_OP(ExitFunction) {
     jmp(qword[LookupBase + 0]);
 
     L(FullLookup);
-    mov(rax, ThreadSharedData.Dispatcher->AbsoluteLoopTopAddress);
     mov(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.rip)], RipReg);
-    jmp(rax);
+    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.DispatcherLoopTop)]);
   }
 
 #ifdef BLOCKSTATS
@@ -187,15 +181,13 @@ DEF_OP(Syscall) {
   }
 
   mov(rsi, STATE); // Move thread in to rsi
-  mov(rdi, reinterpret_cast<uint64_t>(CTX->SyscallHandler));
+  mov(rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SyscallHandlerObj)]);
   mov(rdx, rsp);
 
-  mov(rax, reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall));
-
   if (NumPush & 1)
     sub(rsp, 8); // Align
   // {rdi, rsi, rdx}
-  call(rax);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SyscallHandlerFunc)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -280,9 +272,7 @@ DEF_OP(RemoveCodeEntry) {
   mov(rax, Entry); // imm64 move
   mov(rsi, rax);
 
-
-  mov(rax, reinterpret_cast<uintptr_t>(&Context::Context::RemoveCodeEntryFromJit));
-  call(rax);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.RemoveCodeEntryFromJIT)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -294,13 +284,6 @@ DEF_OP(RemoveCodeEntry) {
 DEF_OP(CPUID) {
   auto Op = IROp->C<IR::IROp_CPUID>();
 
-  using ClassPtrType = FEXCore::CPUID::FunctionResults (FEXCore::CPUIDEmu::*)(uint32_t Function, uint32_t Leaf);
-  union {
-    ClassPtrType ClassPtr;
-    uint64_t Raw;
-  } Ptr;
-  Ptr.ClassPtr = &CPUIDEmu::RunFunction;
-
   for (auto &Reg : RA64)
     push(Reg);
 
@@ -313,18 +296,15 @@ 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, reinterpret_cast<uint64_t>(&CTX->CPUID));
+  mov (rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.CPUIDObj)]);
 
   auto NumPush = RA64.size();
 
   if (NumPush & 1)
     sub(rsp, 8); // Align
 
-  mov(rax, Ptr.Raw);
-
   // {rdi, rsi, rdx}
-
-  call(rax);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.CPUIDFunction)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -342,7 +322,6 @@ 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(GUESTRETURN,       GuestReturn);
   REGISTER_OP(SIGNALRETURN,      SignalReturn);
   REGISTER_OP(CALLBACKRETURN,    CallbackReturn);
   REGISTER_OP(EXITFUNCTION,      ExitFunction);

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

@@ -18,23 +18,23 @@ namespace FEXCore::CPU {
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 DEF_OP(VInsGPR) {
   auto Op = IROp->C<IR::IROp_VInsGPR>();
-  movapd(GetDst(Node), GetSrc(Op->Header.Args[0].ID()));
+  movapd(GetDst(Node), GetSrc(Op->DestVector.ID()));
 
   switch (Op->Header.ElementSize) {
     case 1: {
-      pinsrb(GetDst(Node), GetSrc<RA_32>(Op->Header.Args[1].ID()), Op->Index);
+      pinsrb(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()), Op->DestIdx);
       break;
     }
     case 2: {
-      pinsrw(GetDst(Node), GetSrc<RA_32>(Op->Header.Args[1].ID()), Op->Index);
+      pinsrw(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()), Op->DestIdx);
       break;
     }
     case 4: {
-      pinsrd(GetDst(Node), GetSrc<RA_32>(Op->Header.Args[1].ID()), Op->Index);
+      pinsrd(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()), Op->DestIdx);
       break;
     }
     case 8: {
-      pinsrq(GetDst(Node), GetSrc<RA_64>(Op->Header.Args[1].ID()), Op->Index);
+      pinsrq(GetDst(Node), GetSrc<RA_64>(Op->Src.ID()), Op->DestIdx);
       break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;

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

@@ -45,6 +45,36 @@ DEF_OP(AESKeyGenAssist) {
   vaeskeygenassist(GetDst(Node), GetSrc(Op->Header.Args[0].ID()), Op->RCON);
 }
 
+DEF_OP(CRC32) {
+  auto Op = IROp->C<IR::IROp_CRC32>();
+  switch (IROp->Size) {
+  case 4:
+    mov(TMP1, GetSrc<RA_32>(Op->Header.Args[1].ID()));
+    mov(GetDst<RA_32>(Node), GetSrc<RA_32>(Op->Header.Args[0].ID()));
+  break;
+  case 8:
+    mov(TMP1, GetSrc<RA_64>(Op->Header.Args[1].ID()));
+    mov(GetDst<RA_64>(Node), GetSrc<RA_64>(Op->Header.Args[0].ID()));
+  break;
+  default: LOGMAN_MSG_A_FMT("Unknown CRC32 size: {}", IROp->Size);
+  }
+
+  switch (Op->SrcSize) {
+    case 1:
+      crc32(GetDst<RA_32>(Node).cvt32(), TMP1.cvt8());
+      break;
+    case 2:
+      crc32(GetDst<RA_32>(Node).cvt32(), TMP1.cvt16());
+      break;
+    case 4:
+      crc32(GetDst<RA_32>(Node).cvt32(), TMP1.cvt32());
+      break;
+    case 8:
+      crc32(GetDst<RA_64>(Node).cvt64(), TMP1.cvt64());
+      break;
+  }
+}
+
 #undef DEF_OP
 void X86JITCore::RegisterEncryptionHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
@@ -54,7 +84,7 @@ void X86JITCore::RegisterEncryptionHandlers() {
   REGISTER_OP(VAESDEC,     AESDec);
   REGISTER_OP(VAESDECLAST, AESDecLast);
   REGISTER_OP(VAESKEYGENASSIST, AESKeyGenAssist);
-
+  REGISTER_OP(CRC32,             CRC32);
 #undef REGISTER_OP
 }
 }

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

@@ -15,6 +15,8 @@ $end_info$
 #include "Interface/IR/PassManager.h"
 #include "Interface/IR/Passes/RegisterAllocationPass.h"
 
+#include "Utils/MemberFunctionToPointer.h"
+
 #include <FEXCore/Core/CPUBackend.h>
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Core/SignalDelegator.h>
@@ -27,7 +29,6 @@ $end_info$
 
 #include <algorithm>
 #include <array>
-#include <bits/types/stack_t.h>
 #include <memory>
 #include <stddef.h>
 #include <stdint.h>
@@ -42,6 +43,16 @@ $end_info$
 // #define DEBUG_RA 1
 // #define DEBUG_CYCLES
 
+namespace {
+static void PrintValue(uint64_t Value) {
+  LogMan::Msg::DFmt("Value: 0x{:x}", Value);
+}
+
+static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
+  LogMan::Msg::DFmt("Value: 0x{:016x}'{:016x}", ValueUpper, Value);
+}
+}
+
 namespace FEXCore::CPU {
 
 CodeBuffer AllocateNewCodeBuffer(FEXCore::Context::Context *CTX, size_t Size) {
@@ -109,9 +120,7 @@ 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()));
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
         break;
@@ -120,9 +129,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         movss(xmm0, GetSrc(IROp->Args[0].ID()));
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -136,9 +143,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         movsd(xmm0, GetSrc(IROp->Args[0].ID()));
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -153,9 +158,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         mov(edi, GetSrc<RA_32>(IROp->Args[0].ID()));
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -171,9 +174,7 @@ 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);
 
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -187,9 +188,7 @@ 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);
 
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -203,9 +202,7 @@ 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);
 
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -218,9 +215,7 @@ 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);
 
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -233,9 +228,7 @@ 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);
 
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -251,9 +244,7 @@ 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);
 
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -266,9 +257,7 @@ 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);
 
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -286,9 +275,7 @@ 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);
 
-        mov(rax, (uintptr_t)Info.fn);
-
-        call(rax);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -318,6 +305,7 @@ X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalTh
   , InitialCodeBuffer {Buffer}
 {
   CurrentCodeBuffer = &InitialCodeBuffer;
+  EmitDetectionString();
 
   RAPass = Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA");
 
@@ -350,6 +338,7 @@ X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalTh
     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<X86Dispatcher>(CTX, ThreadState, config);
     DispatchPtr = Dispatcher->DispatchPtr;
@@ -361,26 +350,51 @@ X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalTh
     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);
+  {
+    // 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);
 
-    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);
+      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);
+  }
+}
+
+void X86JITCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
+  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);
   }
 }
 
@@ -394,6 +408,13 @@ X86JITCore::~X86JITCore() {
   FreeCodeBuffer(InitialCodeBuffer);
 }
 
+void X86JITCore::EmitDetectionString() {
+  const char JITString[] = "FEXJIT::X86JITCore::";
+  for (char c : JITString) {
+    db(c);
+  }
+}
+
 void X86JITCore::ClearCache() {
   if (*ThreadSharedData.SignalHandlerRefCounterPtr == 0) {
     if (!CodeBuffers.empty()) {
@@ -432,6 +453,8 @@ void X86JITCore::ClearCache() {
     EmplaceNewCodeBuffer(NewCodeBuffer);
     setNewBuffer(NewCodeBuffer.Ptr, NewCodeBuffer.Size);
   }
+
+  EmitDetectionString();
 }
 
 IR::PhysicalRegister X86JITCore::GetPhys(IR::NodeID Node) const {
@@ -553,7 +576,12 @@ bool X86JITCore::IsInlineEntrypointOffset(const IR::OrderedNodeWrapper& WNode, u
   if (OpHeader->Op == IR::IROps::OP_INLINEENTRYPOINTOFFSET) {
     auto Op = OpHeader->C<IR::IROp_InlineEntrypointOffset>();
     if (Value) {
-      *Value = Entry + Op->Offset;
+      uint64_t Mask = ~0ULL;
+      uint8_t OpSize = OpHeader->Size;
+      if (OpSize == 4) {
+        Mask = 0xFFFF'FFFFULL;
+      }
+      *Value = (Entry + Op->Offset) & Mask;
     }
     return true;
   } else {
@@ -788,4 +816,9 @@ uint64_t X86JITCore::ExitFunctionLink(X86JITCore *core, FEXCore::Core::CpuStateF
 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);
 }
+
+void InitializeX86JITSignalHandlers(FEXCore::Context::Context *CTX) {
+  X86JITCore::InitializeSignalHandlers(CTX);
+}
+
 }

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

@@ -83,6 +83,8 @@ public:
     return Dispatcher->IsAddressInJITCode(Address, IncludeDispatcher, IncludeCompileService);
   }
 
+  static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
+
 private:
   Label* PendingTargetLabel{};
   FEXCore::Context::Context *CTX;
@@ -152,6 +154,9 @@ private:
 
   static uint64_t ExitFunctionLink(X86JITCore* code, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record);
 
+  // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
+  void EmitDetectionString();
+
   // 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
@@ -276,7 +281,6 @@ private:
   ///< Branch ops
   DEF_OP(GuestCallDirect);
   DEF_OP(GuestCallIndirect);
-  DEF_OP(GuestReturn);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -329,6 +333,7 @@ private:
   DEF_OP(GetRoundingMode);
   DEF_OP(SetRoundingMode);
   DEF_OP(ProcessorID);
+  DEF_OP(RDRAND);
 
   ///< Move ops
   DEF_OP(ExtractElementPair);
@@ -338,8 +343,6 @@ private:
   ///< Vector ops
   DEF_OP(VectorZero);
   DEF_OP(VectorImm);
-  DEF_OP(CreateVector2);
-  DEF_OP(CreateVector4);
   DEF_OP(SplatVector);
   DEF_OP(VMov);
   DEF_OP(VAnd);
@@ -426,6 +429,7 @@ private:
   DEF_OP(VSMull2);
   DEF_OP(VUABDL);
   DEF_OP(VTBL1);
+  DEF_OP(VRev64);
 
   ///< Encryption ops
   DEF_OP(AESImc);
@@ -434,6 +438,7 @@ private:
   DEF_OP(AESDec);
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
+  DEF_OP(CRC32);
 #undef DEF_OP
 };
 

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

@@ -23,7 +23,7 @@ DEF_OP(LoadContext) {
   auto Op = IROp->C<IR::IROp_LoadContext>();
   uint8_t OpSize = IROp->Size;
 
-  if (Op->Class.Val == 0) {
+  if (Op->Class == IR::GPRClass) {
     switch (OpSize) {
     case 1: {
       movzx(GetDst<RA_32>(Node), byte [STATE + Op->Offset]);
@@ -84,23 +84,23 @@ DEF_OP(StoreContext) {
   auto Op = IROp->C<IR::IROp_StoreContext>();
   uint8_t OpSize = IROp->Size;
 
-  if (Op->Class.Val == 0) {
+  if (Op->Class == IR::GPRClass) {
     switch (OpSize) {
     case 1: {
-      mov(byte [STATE + Op->Offset], GetSrc<RA_8>(Op->Header.Args[0].ID()));
+      mov(byte [STATE + Op->Offset], GetSrc<RA_8>(Op->Value.ID()));
     }
     break;
 
     case 2: {
-      mov(word [STATE + Op->Offset], GetSrc<RA_16>(Op->Header.Args[0].ID()));
+      mov(word [STATE + Op->Offset], GetSrc<RA_16>(Op->Value.ID()));
     }
     break;
     case 4: {
-      mov(dword [STATE + Op->Offset], GetSrc<RA_32>(Op->Header.Args[0].ID()));
+      mov(dword [STATE + Op->Offset], GetSrc<RA_32>(Op->Value.ID()));
     }
     break;
     case 8: {
-      mov(qword [STATE + Op->Offset], GetSrc<RA_64>(Op->Header.Args[0].ID()));
+      mov(qword [STATE + Op->Offset], GetSrc<RA_64>(Op->Value.ID()));
     }
     break;
     case 16:
@@ -112,27 +112,27 @@ DEF_OP(StoreContext) {
   else {
     switch (OpSize) {
     case 1: {
-      pextrb(byte [STATE + Op->Offset], GetSrc(Op->Header.Args[0].ID()), 0);
+      pextrb(byte [STATE + Op->Offset], GetSrc(Op->Value.ID()), 0);
     }
     break;
 
     case 2: {
-      pextrw(word [STATE + Op->Offset], GetSrc(Op->Header.Args[0].ID()), 0);
+      pextrw(word [STATE + Op->Offset], GetSrc(Op->Value.ID()), 0);
     }
     break;
     case 4: {
-      vmovd(dword [STATE + Op->Offset], GetSrc(Op->Header.Args[0].ID()));
+      vmovd(dword [STATE + Op->Offset], GetSrc(Op->Value.ID()));
     }
     break;
     case 8: {
-      vmovq(qword [STATE + Op->Offset], GetSrc(Op->Header.Args[0].ID()));
+      vmovq(qword [STATE + Op->Offset], GetSrc(Op->Value.ID()));
     }
     break;
     case 16: {
       if (Op->Offset % 16 == 0)
-        movaps(xword [STATE + Op->Offset], GetSrc(Op->Header.Args[0].ID()));
+        movaps(xword [STATE + Op->Offset], GetSrc(Op->Value.ID()));
       else
-        movups(xword [STATE + Op->Offset], GetSrc(Op->Header.Args[0].ID()));
+        movups(xword [STATE + Op->Offset], GetSrc(Op->Value.ID()));
     }
     break;
     default:  LOGMAN_MSG_A_FMT("Unhandled StoreContext size: {}", OpSize);
@@ -143,9 +143,9 @@ DEF_OP(StoreContext) {
 DEF_OP(LoadContextIndexed) {
   auto Op = IROp->C<IR::IROp_LoadContextIndexed>();
   size_t size = IROp->Size;
-  Reg index = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  Reg index = GetSrc<RA_64>(Op->Index.ID());
 
-  if (Op->Class.Val == 0) {
+  if (Op->Class == IR::GPRClass) {
     switch (Op->Stride) {
     case 1:
     case 2:
@@ -245,11 +245,11 @@ DEF_OP(LoadContextIndexed) {
 
 DEF_OP(StoreContextIndexed) {
   auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
-  Reg index = GetSrc<RA_64>(Op->Header.Args[1].ID());
+  Reg index = GetSrc<RA_64>(Op->Index.ID());
   size_t size = IROp->Size;
 
-  if (Op->Class.Val == 0) {
-    auto value = GetSrc<RA_64>(Op->Header.Args[0].ID());
+  if (Op->Class == IR::GPRClass) {
+    auto value = GetSrc<RA_64>(Op->Value.ID());
     lea(rax, dword [STATE + Op->BaseOffset]);
 
     switch (Op->Stride) {
@@ -269,7 +269,7 @@ DEF_OP(StoreContextIndexed) {
     }
   }
   else {
-    auto value = GetSrc(Op->Header.Args[0].ID());
+    auto value = GetSrc(Op->Value.ID());
     switch (Op->Stride) {
     case 1:
     case 2:
@@ -339,19 +339,19 @@ DEF_OP(SpillRegister) {
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {
       case 1: {
-        mov(byte [rsp + SlotOffset], GetSrc<RA_8>(Op->Header.Args[0].ID()));
+        mov(byte [rsp + SlotOffset], GetSrc<RA_8>(Op->Value.ID()));
         break;
       }
       case 2: {
-        mov(word [rsp + SlotOffset], GetSrc<RA_16>(Op->Header.Args[0].ID()));
+        mov(word [rsp + SlotOffset], GetSrc<RA_16>(Op->Value.ID()));
         break;
       }
       case 4: {
-        mov(dword [rsp + SlotOffset], GetSrc<RA_32>(Op->Header.Args[0].ID()));
+        mov(dword [rsp + SlotOffset], GetSrc<RA_32>(Op->Value.ID()));
         break;
       }
       case 8: {
-        mov(qword [rsp + SlotOffset], GetSrc<RA_64>(Op->Header.Args[0].ID()));
+        mov(qword [rsp + SlotOffset], GetSrc<RA_64>(Op->Value.ID()));
         break;
       }
       default:  LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize);
@@ -359,15 +359,15 @@ DEF_OP(SpillRegister) {
   } else if (Op->Class == FEXCore::IR::FPRClass) {
     switch (OpSize) {
       case 4: {
-        movss(dword [rsp + SlotOffset], GetSrc(Op->Header.Args[0].ID()));
+        movss(dword [rsp + SlotOffset], GetSrc(Op->Value.ID()));
         break;
       }
       case 8: {
-        movsd(qword [rsp + SlotOffset], GetSrc(Op->Header.Args[0].ID()));
+        movsd(qword [rsp + SlotOffset], GetSrc(Op->Value.ID()));
         break;
       }
       case 16: {
-        movaps(xword [rsp + SlotOffset], GetSrc(Op->Header.Args[0].ID()));
+        movaps(xword [rsp + SlotOffset], GetSrc(Op->Value.ID()));
         break;
       }
       default:  LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize);
@@ -435,7 +435,7 @@ DEF_OP(LoadFlag) {
 DEF_OP(StoreFlag) {
   auto Op = IROp->C<IR::IROp_StoreFlag>();
 
-  mov (rax, GetSrc<RA_64>(Op->Header.Args[0].ID()));
+  mov (rax, GetSrc<RA_64>(Op->Value.ID()));
   mov(byte [STATE + (offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag)], al);
 }
 
@@ -469,7 +469,7 @@ DEF_OP(LoadMem) {
 
   auto MemPtr = GenerateModRM(MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
 
-  if (Op->Class.Val == 0) {
+  if (Op->Class == IR::GPRClass) {
     auto Dst = GetDst<RA_64>(Node);
 
     switch (IROp->Size) {
@@ -537,19 +537,19 @@ DEF_OP(StoreMem) {
 
   auto MemPtr = GenerateModRM(MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
 
-  if (Op->Class.Val == 0) {
+  if (Op->Class == IR::GPRClass) {
     switch (IROp->Size) {
     case 1:
-      mov(byte [MemPtr], GetSrc<RA_8>(Op->Header.Args[1].ID()));
+      mov(byte [MemPtr], GetSrc<RA_8>(Op->Value.ID()));
     break;
     case 2:
-      mov(word [MemPtr], GetSrc<RA_16>(Op->Header.Args[1].ID()));
+      mov(word [MemPtr], GetSrc<RA_16>(Op->Value.ID()));
     break;
     case 4:
-      mov(dword [MemPtr], GetSrc<RA_32>(Op->Header.Args[1].ID()));
+      mov(dword [MemPtr], GetSrc<RA_32>(Op->Value.ID()));
     break;
     case 8:
-      mov(qword [MemPtr], GetSrc<RA_64>(Op->Header.Args[1].ID()));
+      mov(qword [MemPtr], GetSrc<RA_64>(Op->Value.ID()));
     break;
     default:  LOGMAN_MSG_A_FMT("Unhandled StoreMem size: {}", IROp->Size);
     }
@@ -557,22 +557,22 @@ DEF_OP(StoreMem) {
   else {
     switch (IROp->Size) {
     case 1:
-      pextrb(byte [MemPtr], GetSrc(Op->Header.Args[1].ID()), 0);
+      pextrb(byte [MemPtr], GetSrc(Op->Value.ID()), 0);
     break;
     case 2:
-      pextrw(word [MemPtr], GetSrc(Op->Header.Args[1].ID()), 0);
+      pextrw(word [MemPtr], GetSrc(Op->Value.ID()), 0);
     break;
     case 4:
-      vmovd(dword [MemPtr], GetSrc(Op->Header.Args[1].ID()));
+      vmovd(dword [MemPtr], GetSrc(Op->Value.ID()));
     break;
     case 8:
-      vmovq(qword [MemPtr], GetSrc(Op->Header.Args[1].ID()));
+      vmovq(qword [MemPtr], GetSrc(Op->Value.ID()));
     break;
     case 16:
       if (IROp->Size == Op->Align)
-        movups(xword [MemPtr], GetSrc(Op->Header.Args[1].ID()));
+        movups(xword [MemPtr], GetSrc(Op->Value.ID()));
       else
-        movups(xword [MemPtr], GetSrc(Op->Header.Args[1].ID()));
+        movups(xword [MemPtr], GetSrc(Op->Value.ID()));
     break;
     default:  LOGMAN_MSG_A_FMT("Unhandled StoreMem size: {}", IROp->Size);
     }

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

@@ -18,14 +18,6 @@ $end_info$
 #include <xbyak/xbyak.h>
 
 namespace FEXCore::CPU {
-static void PrintValue(uint64_t Value) {
-  LogMan::Msg::DFmt("Value: 0x{:x}", Value);
-}
-
-static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
-  LogMan::Msg::DFmt("Value: 0x{:016x}'{:016x}", ValueUpper, Value);
-}
-
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 
 DEF_OP(Fence) {
@@ -53,8 +45,7 @@ DEF_OP(Break) {
       break;
     case FEXCore::IR::Break_Overflow: // overflow
       // Need to be outside of JIT cache space to ensure cache clearing correctness
-      mov(TMP1, ThreadSharedData.Dispatcher->OverflowExceptionInstructionAddress);
-      jmp(TMP1);
+      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.OverflowExceptionHandler)]);
       break;
     case FEXCore::IR::Break_Halt: { // HLT
       // Time to quit
@@ -62,8 +53,7 @@ DEF_OP(Break) {
       mov(rsp, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)]);
 
       // Now we need to jump to the thread stop handler
-      mov(TMP1, ThreadSharedData.Dispatcher->ThreadStopHandlerAddress);
-      jmp(TMP1);
+      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadStopHandler)]);
       break;
     }
     case FEXCore::IR::Break_Interrupt3: // INT3
@@ -75,8 +65,7 @@ DEF_OP(Break) {
         }
 
         // This jump target needs to be a constant offset here
-        mov(TMP1, ThreadSharedData.Dispatcher->ThreadPauseHandlerAddress);
-        jmp(TMP1);
+        jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadPauseHandler)]);
       }
       else {
         // If we don't have a gdb server attached then....crash?
@@ -84,8 +73,7 @@ DEF_OP(Break) {
         mov(rsp, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)]);
 
         // Now we need to jump to the thread stop handler
-        mov(TMP1, ThreadSharedData.Dispatcher->ThreadStopHandlerAddress);
-        jmp(TMP1);
+        jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadStopHandler)]);
       }
     break;
     }
@@ -96,9 +84,7 @@ DEF_OP(Break) {
       }
 
       // Need to be outside of JIT cache space to ensure cache clearing correctness
-      mov(TMP1, ThreadSharedData.Dispatcher->UnimplementedInstructionAddress);
-      jmp(TMP1);
-
+      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.UnimplementedInstructionHandler)]);
       break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown Break reason: {}", Op->Reason);
@@ -144,18 +130,15 @@ DEF_OP(Print) {
   PushRegs();
   if (IsGPR(Op->Header.Args[0].ID())) {
     mov (rdi, GetSrc<RA_64>(Op->Header.Args[0].ID()));
-
-    mov(rax, reinterpret_cast<uintptr_t>(PrintValue));
+    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.PrintValue)]);
   }
   else {
     pextrq(rdi, GetSrc(Op->Header.Args[0].ID()), 0);
     pextrq(rsi, GetSrc(Op->Header.Args[0].ID()), 1);
 
-    mov(rax, reinterpret_cast<uintptr_t>(PrintVectorValue));
+    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.PrintVectorValue)]);
   }
 
-  call(rax);
-
   PopRegs();
 }
 
@@ -166,6 +149,23 @@ DEF_OP(ProcessorID) {
   mov (GetDst<RA_32>(Node), ecx);
 }
 
+DEF_OP(RDRAND) {
+  auto Op = IROp->C<IR::IROp_RDRAND>();
+
+  auto Dst = GetSrcPair<RA_64>(Node);
+
+  if (Op->GetReseeded) {
+    rdrand(Dst.first);
+  }
+  else {
+    rdseed(Dst.first);
+  }
+
+  // In the case of RDRAND or RDSEED returning a valid number then CF = 1, else 0
+  mov (Dst.second, 0);
+  setc(Dst.second.cvt8());
+}
+
 #undef DEF_OP
 void X86JITCore::RegisterMiscHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
@@ -183,6 +183,7 @@ void X86JITCore::RegisterMiscHandlers() {
   REGISTER_OP(SETROUNDINGMODE, SetRoundingMode);
   REGISTER_OP(INVALIDATEFLAGS,   NoOp);
   REGISTER_OP(PROCESSORID,   ProcessorID);
+  REGISTER_OP(RDRAND, RDRAND);
 #undef REGISTER_OP
 }
 }

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

@@ -42,7 +42,7 @@ DEF_OP(CreateElementPair) {
   Xbyak::Reg RegSecond;
   Xbyak::Reg RegTmp;
 
-  switch (Op->Header.Size) {
+  switch (IROp->ElementSize) {
     case 4: {
       Dst = GetSrcPair<RA_32>(Node);
       RegFirst = GetSrc<RA_32>(Op->Header.Args[0].ID());

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

@@ -67,14 +67,6 @@ DEF_OP(VectorImm) {
   }
 }
 
-DEF_OP(CreateVector2) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
-DEF_OP(CreateVector4) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
 DEF_OP(SplatVector) {
   auto Op = IROp->C<IR::IROp_SplatVector2>();
   uint8_t OpSize = IROp->Size;
@@ -1545,7 +1537,7 @@ DEF_OP(VInsScalarElement) {
 DEF_OP(VExtractElement) {
   auto Op = IROp->C<IR::IROp_VExtractElement>();
 
-  switch (Op->Header.ElementSize) {
+  switch (Op->Header.Size) {
     case 1: {
       pextrb(eax, GetSrc(Op->Header.Args[0].ID()), Op->Index);
       pinsrb(GetDst(Node), eax, 0);
@@ -1566,7 +1558,7 @@ DEF_OP(VExtractElement) {
       pinsrq(GetDst(Node), rax, 0);
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.Size); break;
   }
 }
 
@@ -2151,13 +2143,79 @@ DEF_OP(VTBL1) {
   }
 }
 
+DEF_OP(VRev64) {
+  auto Op = IROp->C<IR::IROp_VDupElement>();
+
+  switch (Op->Header.ElementSize) {
+    case 1: {
+      mov(rax, 0x00'01'02'03'04'05'06'07); // Lower
+      vmovq(xmm15, rax);
+      if (IROp->Size == 16) {
+        // Full 8bit byteswap in each 64-bit element
+        mov(rcx, 0x08'09'0A'0B'0C'0D'0E'0F); // Upper
+        pinsrq(xmm15, rcx, 1);
+      }
+      else {
+        // 8byte, upper bits get zero
+        // Full 8bit byteswap in each 64-bit element
+        mov(rcx, 0x80'80'80'80'80'80'80'80); // Upper
+        pinsrq(xmm15, rcx, 1);
+      }
+
+      vpshufb(GetDst(Node), GetSrc(Op->Header.Args[0].ID()), xmm15);
+      break;
+    }
+    case 2: {
+      // Full 16-bit byteswap in each 64-bit element
+      mov(rax, 0x01'00'03'02'05'04'07'06); // Lower
+      vmovq(xmm15, rax);
+      if (IROp->Size == 16) {
+        mov(rcx, 0x09'08'0B'0A'0D'0C'0F'0E); // Upper
+        pinsrq(xmm15, rcx, 1);
+      }
+      else {
+        // 8byte, upper bits get zero
+        // Full 8bit byteswap in each 64-bit element
+        mov(rcx, 0x80'80'80'80'80'80'80'80); // Upper
+        pinsrq(xmm15, rcx, 1);
+      }
+      vpshufb(GetDst(Node), GetSrc(Op->Header.Args[0].ID()), xmm15);
+      break;
+    }
+    case 4: {
+      if (IROp->Size == 16) {
+      vpshufd(GetDst(Node),
+        GetSrc(Op->Header.Args[0].ID()),
+        (0b11 << 0) |
+        (0b10 << 2) |
+        (0b01 << 4) |
+        (0b00 << 6));
+      }
+      else {
+
+      vpshufd(GetDst(Node),
+        GetSrc(Op->Header.Args[0].ID()),
+        (0b01 << 0) |
+        (0b00 << 2) |
+        (0b11 << 4) | // Last two don't matter, will be overwritten with zero
+        (0b11 << 6));
+
+        // Zero upper 64-bits
+        mov(rcx, 0);
+        pinsrq(GetDst(Node), rcx, 1);
+      }
+      break;
+    }
+    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+  }
+}
+
+
 #undef DEF_OP
 void X86JITCore::RegisterVectorHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
   REGISTER_OP(VECTORZERO,        VectorZero);
   REGISTER_OP(VECTORIMM,         VectorImm);
-  REGISTER_OP(CREATEVECTOR2,     CreateVector2);
-  REGISTER_OP(CREATEVECTOR4,     CreateVector4);
   REGISTER_OP(SPLATVECTOR2,      SplatVector);
   REGISTER_OP(SPLATVECTOR4,      SplatVector);
   REGISTER_OP(VMOV,              VMov);
@@ -2246,6 +2304,7 @@ void X86JITCore::RegisterVectorHandlers() {
   REGISTER_OP(VSMULL2,           VSMull2);
   REGISTER_OP(VUABDL,            VUABDL);
   REGISTER_OP(VTBL1,             VTBL1);
+  REGISTER_OP(VREV64,            VRev64);
 #undef REGISTER_OP
 }
 }

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

@@ -14,6 +14,7 @@
 #include <FEXCore/Utils/LogManager.h>
 
 #include <cstdint>
+#include <fmt/format.h>
 #include <map>
 #include <stddef.h>
 #include <utility>
@@ -35,6 +36,38 @@ enum class SelectionFlag {
 };
 
 public:
+  enum class FlagsGenerationType : uint8_t {
+    TYPE_NONE,
+    TYPE_ADC,
+    TYPE_SBB,
+    TYPE_SUB,
+    TYPE_ADD,
+    TYPE_MUL,
+    TYPE_UMUL,
+    TYPE_LOGICAL,
+    TYPE_LSHL,
+    TYPE_LSHLI,
+    TYPE_LSHR,
+    TYPE_LSHRI,
+    TYPE_ASHR,
+    TYPE_ASHRI,
+    TYPE_ROR,
+    TYPE_RORI,
+    TYPE_ROL,
+    TYPE_ROLI,
+    TYPE_FCMP,
+    TYPE_BEXTR,
+    TYPE_BLSI,
+    TYPE_BLSMSK,
+    TYPE_BLSR,
+    TYPE_POPCOUNT,
+    TYPE_BZHI,
+    TYPE_TZCNT,
+    TYPE_LZCNT,
+    TYPE_BITSELECT,
+    TYPE_RDRAND,
+  };
+
   SelectionFlag flagsOp{};
   uint8_t flagsOpSize{};
   OrderedNode* flagsOpDest{};
@@ -87,6 +120,9 @@ public:
     //  cmp qword [rdi-8], 0
     //  jne .label
     if (LastOp && !BlockSetRIP) {
+      // Calculate flags first
+      CalculateDeferredFlags();
+
       auto it = JumpTargets.find(NextRIP);
       if (it == JumpTargets.end()) {
 
@@ -101,6 +137,11 @@ public:
         return true;
       }
     }
+
+    if (LastOp) {
+      LOGMAN_THROW_A_FMT(IsDeferredFlagsStored(), "FinishOp: Deferred flags weren't generated at end of block");
+    }
+
     BlockSetRIP = false;
 
     return false;
@@ -233,6 +274,8 @@ public:
   void XADDOp(OpcodeArgs);
   void PopcountOp(OpcodeArgs);
   void XLATOp(OpcodeArgs);
+  template<bool Reseed>
+  void RDRANDOp(OpcodeArgs);
 
   enum class Segment {
     FS,
@@ -256,9 +299,14 @@ public:
   template<FEXCore::IR::IROps IROp, size_t ElementSize>
   void VectorALUOp(OpcodeArgs);
   template<FEXCore::IR::IROps IROp, size_t ElementSize>
+  void VectorALUROp(OpcodeArgs);
+  template<FEXCore::IR::IROps IROp, size_t ElementSize>
   void VectorScalarALUOp(OpcodeArgs);
   template<FEXCore::IR::IROps IROp, size_t ElementSize, bool Scalar>
   void VectorUnaryOp(OpcodeArgs);
+  template<FEXCore::IR::IROps IROp, size_t ElementSize>
+  void VectorUnaryDuplicateOp(OpcodeArgs);
+
   void MOVQOp(OpcodeArgs);
   template<size_t ElementSize>
   void PADDQOp(OpcodeArgs);
@@ -445,6 +493,17 @@ public:
   template<size_t ElementSize>
   void ADDSUBPOp(OpcodeArgs);
 
+  void PFNACCOp(OpcodeArgs);
+  void PFPNACCOp(OpcodeArgs);
+  void PSWAPDOp(OpcodeArgs);
+
+  template<uint8_t CompType>
+  void VPFCMPOp(OpcodeArgs);
+  void PI2FWOp(OpcodeArgs);
+  void PF2IWOp(OpcodeArgs);
+
+  void PMULHRWOp(OpcodeArgs);
+
   void PMADDWD(OpcodeArgs);
   void PMADDUBSW(OpcodeArgs);
 
@@ -500,6 +559,8 @@ public:
 
   void MPSADBWOp(OpcodeArgs);
 
+  void CRC32(OpcodeArgs);
+
   void UnimplementedOp(OpcodeArgs);
 
   void InvalidOp(OpcodeArgs);
@@ -519,7 +580,7 @@ private:
 
   OrderedNode *AppendSegmentOffset(OrderedNode *Value, uint32_t Flags, uint32_t DefaultPrefix = 0, bool Override = false);
 
-  OrderedNode *GetDynamicPC(FEXCore::X86Tables::DecodedOp const& Op, int64_t Offset = 0);
+  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);
@@ -557,23 +618,524 @@ private:
 
   OrderedNode *SelectCC(uint8_t OP, OrderedNode *TrueValue, OrderedNode *FalseValue);
 
-  void GenerateFlags_ADC(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF);
-  void GenerateFlags_SBB(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF);
-  void GenerateFlags_SUB(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF = true);
-  void GenerateFlags_ADD(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF = true);
-  void GenerateFlags_MUL(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *High);
-  void GenerateFlags_UMUL(FEXCore::X86Tables::DecodedOp Op, OrderedNode *High);
-  void GenerateFlags_Logical(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
-  void GenerateFlags_ShiftLeft(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
-  void GenerateFlags_ShiftLeftImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
-  void GenerateFlags_ShiftRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
-  void GenerateFlags_ShiftRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
-  void GenerateFlags_SignShiftRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
-  void GenerateFlags_SignShiftRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
-  void GenerateFlags_RotateRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
-  void GenerateFlags_RotateLeft(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
-  void GenerateFlags_RotateRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
-  void GenerateFlags_RotateLeftImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
+  /**
+   * @name Deferred RFLAG calculation and generation.
+   *
+   * Only handles the six flags that ALU ops typically generate.
+   * Specifically: CF, PF, AF, ZF, SF, OF
+   *  These six flags are heavily generated through basic ALU ops and balloon the IR if not early eliminated.
+   *  This tracking structure only tracks single blocks and requires RFLAGS calculation at block-ending ops.
+   *  Some flags generating ALU ops only touch part of the registers, In these cases it will do calculation up front.
+   *  This means we still need our IR passes to eliminate all redundant flags accesses but this light OpcodeDispatcher optimization
+   *  doesn't take it to that level.
+   * @{ */
+
+  // Deferred flag generation tracking structure.
+  // This structure is used to track RFlags from ALU ops for invalidation.
+  //
+  // Future ideas: Use an invalidation mask to do partial generation of flags.
+  // Particularly for the instructions that don't do the full set of flags calculations.
+  // These instructions currently calculate the deferred RFLAGS immediately then overwrite rflags state.
+  // RCLSE IR pass will catch and remove redundant rflags stores like this currently.
+  struct DeferredFlagData {
+    // What type of flags to generate
+    FlagsGenerationType Type {FlagsGenerationType::TYPE_NONE};
+
+    // Source size of the op
+    uint8_t SrcSize;
+
+    // Every flag generation type has a result
+    OrderedNode *Res{};
+
+    union {
+      // UMUL, BEXTR, BLSI, BLSMSK, POPCOUNT, TZCNT, LZCNT, BITSELECT, RDRAND
+      struct {
+      } NoSource;
+
+      // MUL, BLSR, BZHI
+      struct {
+        OrderedNode *Src1;
+      } OneSource;
+
+      // Logical, LSHL, LSHR, ASHR, ROR, ROL
+      struct {
+        OrderedNode *Src1;
+        OrderedNode *Src2;
+      } TwoSource;
+
+      // ADC, SBB
+      struct {
+        OrderedNode *Src1;
+        OrderedNode *Src2;
+        OrderedNode *Src3;
+      } ThreeSource;
+
+      // LSHLI, LSHRI, ASHRI, RORI, ROLI
+      struct {
+        OrderedNode *Src1;
+        uint64_t Imm;
+      } OneSrcImmediate;
+
+      // ADD, SUB
+      struct {
+        OrderedNode *Src1;
+        OrderedNode *Src2;
+
+        bool UpdateCF;
+      } TwoSrcImmediate;
+    } Sources{};
+  };
+
+  DeferredFlagData CurrentDeferredFlags{};
+
+  /**
+   * @brief Takes the current deferred flag state and stores the result in to RFLAGS.
+   *
+   * Once executed there will no longer be any deferred flag state and RFLAGS will have the correct flags in it.
+   * Necessary to do when leaving a IR block, or if an instruction is doing a partial overwrite of the flags.
+   */
+  void CalculateDeferredFlags(uint32_t FlagsToCalculateMask = ~0U);
+
+  /**
+   * @brief Invalidates the current deferred flags structure.
+   *
+   * If the emulated instruction is going to overwrite all of the flags but isn't tracked using the deferred flag system
+   * then use this function to stop tracking the current active deferred flags.
+   */
+  void InvalidateDeferredFlags() {
+    CurrentDeferredFlags.Type = FlagsGenerationType::TYPE_NONE;
+  }
+
+  /**
+   * @brief Checks if there is any deferred flag state active.
+   *
+   * @return True if RFLAGs contains the flags. False if deferred flags is tracking the data.
+   */
+  bool IsDeferredFlagsStored() const {
+    return CurrentDeferredFlags.Type == FlagsGenerationType::TYPE_NONE;
+  }
+
+  /**
+   * @name These functions are used by the deferred flag handling while it is calculating and storing flags in to RFLAGs.
+   * @{ */
+  void CalculcateFlags_ADC(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF);
+  void CalculcateFlags_SBB(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF);
+  void CalculcateFlags_SUB(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF = true);
+  void CalculcateFlags_ADD(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF = true);
+  void CalculcateFlags_MUL(uint8_t SrcSize, OrderedNode *Res, OrderedNode *High);
+  void CalculcateFlags_UMUL(OrderedNode *High);
+  void CalculcateFlags_Logical(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
+  void CalculcateFlags_ShiftLeft(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
+  void CalculcateFlags_ShiftLeftImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
+  void CalculcateFlags_ShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
+  void CalculcateFlags_ShiftRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
+  void CalculcateFlags_SignShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
+  void CalculcateFlags_SignShiftRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
+  void CalculcateFlags_RotateRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
+  void CalculcateFlags_RotateLeft(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
+  void CalculcateFlags_RotateRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
+  void CalculcateFlags_RotateLeftImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
+  void CalculcateFlags_FCMP(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
+  void CalculcateFlags_BEXTR(OrderedNode *Src);
+  void CalculcateFlags_BLSI(uint8_t SrcSize, OrderedNode *Src);
+  void CalculcateFlags_BLSMSK(OrderedNode *Src);
+  void CalculcateFlags_BLSR(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src);
+  void CalculcateFlags_POPCOUNT(OrderedNode *Src);
+  void CalculcateFlags_BZHI(uint8_t SrcSize, OrderedNode *Result, OrderedNode *Src);
+  void CalculcateFlags_TZCNT(OrderedNode *Src);
+  void CalculcateFlags_LZCNT(uint8_t SrcSize, OrderedNode *Src);
+  void CalculcateFlags_BITSELECT(OrderedNode *Src);
+  void CalculcateFlags_RDRAND(OrderedNode *Src);
+  /**  @} */
+
+  /**
+   * @name These functions generated deferred RFLAGs tracking.
+   *
+   * Depending on the operation it may force a RFLAGs calculation before storing the new deferred state.
+   * @{ */
+  void GenerateFlags_ADC(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_ADC,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .ThreeSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+          .Src3 = CF,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_SBB(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_SBB,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .ThreeSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+          .Src3 = CF,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_SUB(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF = true) {
+    if (!UpdateCF) {
+      // If we aren't updating CF then we need to calculate flags. Invalidation mask would make this not required.
+      CalculateDeferredFlags();
+    }
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_SUB,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSrcImmediate = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+          .UpdateCF = UpdateCF,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_ADD(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF = true) {
+    if (!UpdateCF) {
+      // If we aren't updating CF then we need to calculate flags. Invalidation mask would make this not required.
+      CalculateDeferredFlags();
+    }
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_ADD,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSrcImmediate = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+          .UpdateCF = UpdateCF,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_MUL(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *High) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_MUL,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .OneSource = {
+          .Src1 = High,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_UMUL(FEXCore::X86Tables::DecodedOp Op, OrderedNode *High) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_UMUL,
+      .SrcSize = GetSrcSize(Op),
+      .Res = High,
+    };
+  }
+
+  void GenerateFlags_Logical(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_LOGICAL,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_ShiftLeft(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+    // Flags need to be used, generate incoming flags first.
+    CalculateDeferredFlags();
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_LSHL,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_ShiftRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+    // Flags need to be used, generate incoming flags first.
+    CalculateDeferredFlags();
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_LSHR,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_SignShiftRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+    // Flags need to be used, generate incoming flags first.
+    CalculateDeferredFlags();
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_ASHR,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_ShiftLeftImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+    // No flags changed if shift is zero.
+    if (Shift == 0) return;
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_LSHLI,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .OneSrcImmediate = {
+          .Src1 = Src1,
+          .Imm = Shift,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_SignShiftRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+    // No flags changed if shift is zero.
+    if (Shift == 0) return;
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_ASHRI,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .OneSrcImmediate = {
+          .Src1 = Src1,
+          .Imm = Shift,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_ShiftRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+    // No flags changed if shift is zero.
+    if (Shift == 0) return;
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_LSHRI,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .OneSrcImmediate = {
+          .Src1 = Src1,
+          .Imm = Shift,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_RotateRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+    // Doesn't set all the flags, needs to calculate.
+    CalculateDeferredFlags();
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_ROR,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_RotateLeft(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+    // Doesn't set all the flags, needs to calculate.
+    CalculateDeferredFlags();
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_ROL,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_RotateRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+    if (Shift == 0) return;
+
+    // Doesn't set all the flags, needs to calculate.
+    CalculateDeferredFlags();
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_RORI,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .OneSrcImmediate = {
+          .Src1 = Src1,
+          .Imm = Shift,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_RotateLeftImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+    if (Shift == 0) return;
+
+    // Doesn't set all the flags, needs to calculate.
+    CalculateDeferredFlags();
+
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_ROLI,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .OneSrcImmediate = {
+          .Src1 = Src1,
+          .Imm = Shift,
+        },
+      }
+    };
+  }
+
+  void GenerateFlags_FCMP(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_FCMP,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .TwoSource = {
+          .Src1 = Src1,
+          .Src2 = Src2,
+        },
+      }
+    };
+  }
+
+  void GenerateFlags_BEXTR(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_BEXTR,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Src,
+    };
+  }
+
+  void GenerateFlags_BLSI(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_BLSI,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Src,
+    };
+  }
+
+  void GenerateFlags_BLSMSK(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_BLSMSK,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Src,
+    };
+  }
+
+  void GenerateFlags_BLSR(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_BLSR,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Res,
+      .Sources = {
+        .OneSource = {
+          .Src1 = Src,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_POPCOUNT(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_POPCOUNT,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Src,
+    };
+  }
+
+  void GenerateFlags_BZHI(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Result, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_BZHI,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Result,
+      .Sources = {
+        .OneSource = {
+          .Src1 = Src,
+        },
+      },
+    };
+  }
+
+  void GenerateFlags_TZCNT(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_TZCNT,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Src,
+    };
+  }
+
+  void GenerateFlags_LZCNT(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_LZCNT,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Src,
+    };
+  }
+
+  void GenerateFlags_BITSELECT(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_BITSELECT,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Src,
+    };
+  }
+
+  void GenerateFlags_RDRAND(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
+    CurrentDeferredFlags = DeferredFlagData {
+      .Type = FlagsGenerationType::TYPE_RDRAND,
+      .SrcSize = GetSrcSize(Op),
+      .Res = Src,
+    };
+  }
+
+  /**  @} */
+  /**  @} */
 
   OrderedNode * GetX87Top();
   enum class X87Tag {
@@ -600,24 +1162,37 @@ private:
   bool Multiblock{};
   uint64_t Entry;
 
-  OrderedNode* _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, uint8_t Size, OrderedNode *ssa0, OrderedNode *ssa1, uint8_t Align = 1) {
+  OrderedNode* _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, uint8_t Size, OrderedNode *Addr, OrderedNode *Value, uint8_t Align = 1) {
     if (CTX->Config.TSOEnabled)
-    	return _StoreMemTSO(ssa0, ssa1, Invalid(), Align, Class, MEM_OFFSET_SXTX, 1, Size);
+      return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
     else
-      return _StoreMem(ssa0, ssa1, Invalid(), Align, Class, MEM_OFFSET_SXTX, 1, Size);
+      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)
-      return _LoadMemTSO(ssa0, Invalid(), Align, Class, MEM_OFFSET_SXTX, 1, Size);
+      return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
     else
-      return _LoadMem(ssa0, Invalid(), Align, Class, MEM_OFFSET_SXTX, 1, Size);
+      return _LoadMem(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
   }
 
-
+  void InstallHostSpecificOpcodeHandlers();
 };
 
 void InstallOpcodeHandlers(Context::OperatingMode Mode);
 
 }
+template <>
+struct fmt::formatter<FEXCore::IR::OpDispatchBuilder::FlagsGenerationType> : fmt::formatter<int> {
+  using Base = fmt::formatter<int>;
+
+  // Pass-through the underlying value, so IDs can
+  // be formatted like any integral value.
+  template <typename FormatContext>
+  auto format(const FEXCore::IR::OpDispatchBuilder::FlagsGenerationType& ID, FormatContext& ctx) {
+    return Base::format(static_cast<int>(ID), ctx);
+  }
+};
+
+
 

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

@@ -41,8 +41,17 @@ constexpr std::array<uint32_t, 17> FlagOffsets = {
 void OpDispatchBuilder::SetPackedRFLAG(bool Lower8, OrderedNode *Src) {
   size_t NumFlags = FlagOffsets.size();
   if (Lower8) {
+    // Calculate flags early.
+    // Could use InvalidateDeferredFlags() if we had masked invalidation.
+    // This is only a partial overwrite of flags since OF isn't stored here.
+    CalculateDeferredFlags();
     NumFlags = 5;
   }
+  else {
+    // We are overwriting all RFLAGS. Invalidate the deferred flag state.
+    InvalidateDeferredFlags();
+  }
+
   auto OneConst = _Constant(1);
   for (size_t i = 0; i < NumFlags; ++i) {
     const auto FlagOffset = FlagOffsets[i];
@@ -52,6 +61,9 @@ void OpDispatchBuilder::SetPackedRFLAG(bool Lower8, OrderedNode *Src) {
 }
 
 OrderedNode *OpDispatchBuilder::GetPackedRFLAG(bool Lower8) {
+  // Calculate flags early.
+  CalculateDeferredFlags();
+
   OrderedNode *Original = _Constant(2);
   size_t NumFlags = FlagOffsets.size();
   if (Lower8) {
@@ -68,8 +80,188 @@ OrderedNode *OpDispatchBuilder::GetPackedRFLAG(bool Lower8) {
   return Original;
 }
 
-void OpDispatchBuilder::GenerateFlags_ADC(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF) {
-  auto Size = GetSrcSize(Op) * 8;
+void OpDispatchBuilder::CalculateDeferredFlags(uint32_t FlagsToCalculateMask) {
+  if (CurrentDeferredFlags.Type == FlagsGenerationType::TYPE_NONE) {
+    // Nothing to do
+    return;
+  }
+
+  switch (CurrentDeferredFlags.Type) {
+    case FlagsGenerationType::TYPE_ADC:
+      CalculcateFlags_ADC(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.ThreeSource.Src1,
+        CurrentDeferredFlags.Sources.ThreeSource.Src2,
+        CurrentDeferredFlags.Sources.ThreeSource.Src3);
+      break;
+    case FlagsGenerationType::TYPE_SBB:
+      CalculcateFlags_SBB(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.ThreeSource.Src1,
+        CurrentDeferredFlags.Sources.ThreeSource.Src2,
+        CurrentDeferredFlags.Sources.ThreeSource.Src3);
+      break;
+    case FlagsGenerationType::TYPE_SUB:
+      CalculcateFlags_SUB(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSrcImmediate.Src1,
+        CurrentDeferredFlags.Sources.TwoSrcImmediate.Src2,
+        CurrentDeferredFlags.Sources.TwoSrcImmediate.UpdateCF);
+      break;
+    case FlagsGenerationType::TYPE_ADD:
+      CalculcateFlags_ADD(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSrcImmediate.Src1,
+        CurrentDeferredFlags.Sources.TwoSrcImmediate.Src2,
+        CurrentDeferredFlags.Sources.TwoSrcImmediate.UpdateCF);
+      break;
+    case FlagsGenerationType::TYPE_MUL:
+      CalculcateFlags_MUL(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.OneSource.Src1);
+      break;
+    case FlagsGenerationType::TYPE_UMUL:
+      CalculcateFlags_UMUL(CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_LOGICAL:
+      CalculcateFlags_Logical(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSource.Src1,
+        CurrentDeferredFlags.Sources.TwoSource.Src2);
+      break;
+    case FlagsGenerationType::TYPE_LSHL:
+      CalculcateFlags_ShiftLeft(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSource.Src1,
+        CurrentDeferredFlags.Sources.TwoSource.Src2);
+      break;
+    case FlagsGenerationType::TYPE_LSHLI:
+      CalculcateFlags_ShiftLeftImmediate(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Src1,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Imm);
+      break;
+    case FlagsGenerationType::TYPE_LSHR:
+      CalculcateFlags_ShiftRight(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSource.Src1,
+        CurrentDeferredFlags.Sources.TwoSource.Src2);
+      break;
+    case FlagsGenerationType::TYPE_LSHRI:
+      CalculcateFlags_ShiftRightImmediate(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Src1,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Imm);
+      break;
+    case FlagsGenerationType::TYPE_ASHR:
+      CalculcateFlags_SignShiftRight(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSource.Src1,
+        CurrentDeferredFlags.Sources.TwoSource.Src2);
+      break;
+    case FlagsGenerationType::TYPE_ASHRI:
+      CalculcateFlags_SignShiftRightImmediate(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Src1,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Imm);
+      break;
+    case FlagsGenerationType::TYPE_ROR:
+      CalculcateFlags_RotateRight(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSource.Src1,
+        CurrentDeferredFlags.Sources.TwoSource.Src2);
+      break;
+    case FlagsGenerationType::TYPE_RORI:
+      CalculcateFlags_RotateRightImmediate(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Src1,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Imm);
+      break;
+    case FlagsGenerationType::TYPE_ROL:
+      CalculcateFlags_RotateLeft(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSource.Src1,
+        CurrentDeferredFlags.Sources.TwoSource.Src2);
+      break;
+    case FlagsGenerationType::TYPE_ROLI:
+      CalculcateFlags_RotateLeftImmediate(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Src1,
+        CurrentDeferredFlags.Sources.OneSrcImmediate.Imm);
+      break;
+    case FlagsGenerationType::TYPE_FCMP:
+      CalculcateFlags_FCMP(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.TwoSource.Src1,
+        CurrentDeferredFlags.Sources.TwoSource.Src2);
+      break;
+    case FlagsGenerationType::TYPE_BEXTR:
+      CalculcateFlags_BEXTR(CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_BLSI:
+      CalculcateFlags_BLSI(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_BLSMSK:
+      CalculcateFlags_BLSMSK(CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_BLSR:
+      CalculcateFlags_BLSR(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.OneSource.Src1);
+      break;
+    case FlagsGenerationType::TYPE_POPCOUNT:
+      CalculcateFlags_POPCOUNT(CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_BZHI:
+      CalculcateFlags_BZHI(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res,
+        CurrentDeferredFlags.Sources.OneSource.Src1);
+      break;
+    case FlagsGenerationType::TYPE_TZCNT:
+      CalculcateFlags_TZCNT(CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_LZCNT:
+      CalculcateFlags_LZCNT(
+        CurrentDeferredFlags.SrcSize,
+        CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_BITSELECT:
+      CalculcateFlags_BITSELECT(CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_RDRAND:
+      CalculcateFlags_RDRAND(CurrentDeferredFlags.Res);
+      break;
+    case FlagsGenerationType::TYPE_NONE:
+    default: ERROR_AND_DIE_FMT("Unhandled flags type {}", CurrentDeferredFlags.Type);
+  }
+
+  // Done calculating
+  CurrentDeferredFlags.Type = FlagsGenerationType::TYPE_NONE;
+}
+
+void OpDispatchBuilder::CalculcateFlags_ADC(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF) {
+  auto Size = SrcSize * 8;
   // AF
   {
     OrderedNode *AFRes = _Xor(_Xor(Src1, Src2), Res);
@@ -79,7 +271,7 @@ void OpDispatchBuilder::GenerateFlags_ADC(FEXCore::X86Tables::DecodedOp Op, Orde
 
   // SF
   {
-    auto SignBitConst = _Constant(GetSrcSize(Op) * 8 - 1);
+    auto SignBitConst = _Constant(Size - 1);
 
     auto LshrOp = _Lshr(Res, SignBitConst);
     SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(LshrOp);
@@ -140,9 +332,7 @@ void OpDispatchBuilder::GenerateFlags_ADC(FEXCore::X86Tables::DecodedOp Op, Orde
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_SBB(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF) {
-  const auto SrcSize = GetSrcSize(Op);
-
+void OpDispatchBuilder::CalculcateFlags_SBB(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, OrderedNode *CF) {
   // AF
   {
     OrderedNode *AFRes = _Xor(_Xor(Src1, Src2), Res);
@@ -212,7 +402,7 @@ void OpDispatchBuilder::GenerateFlags_SBB(FEXCore::X86Tables::DecodedOp Op, Orde
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_SUB(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF) {
+void OpDispatchBuilder::CalculcateFlags_SUB(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF) {
   // AF
   {
     OrderedNode *AFRes = _Xor(_Xor(Src1, Src2), Res);
@@ -222,7 +412,7 @@ void OpDispatchBuilder::GenerateFlags_SUB(FEXCore::X86Tables::DecodedOp Op, Orde
 
   // SF
   {
-    auto SignBitConst = _Constant(GetSrcSize(Op) * 8 - 1);
+    auto SignBitConst = _Constant(SrcSize * 8 - 1);
 
     auto LshrOp = _Lshr(Res, SignBitConst);
     SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(LshrOp);
@@ -263,15 +453,13 @@ void OpDispatchBuilder::GenerateFlags_SUB(FEXCore::X86Tables::DecodedOp Op, Orde
     auto XorOp2 = _Xor(Res, Src1);
     OrderedNode *FinalAnd = _And(XorOp1, XorOp2);
 
-    FinalAnd = _Bfe(1, GetSrcSize(Op) * 8 - 1, FinalAnd);
+    FinalAnd = _Bfe(1, SrcSize * 8 - 1, FinalAnd);
 
     SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(FinalAnd);
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_ADD(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF) {
-  const auto SrcSize = GetSrcSize(Op);
-
+void OpDispatchBuilder::CalculcateFlags_ADD(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool UpdateCF) {
   // AF
   {
     OrderedNode *AFRes = _Xor(_Xor(Src1, Src2), Res);
@@ -339,7 +527,7 @@ void OpDispatchBuilder::GenerateFlags_ADD(FEXCore::X86Tables::DecodedOp Op, Orde
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_MUL(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *High) {
+void OpDispatchBuilder::CalculcateFlags_MUL(uint8_t SrcSize, OrderedNode *Res, OrderedNode *High) {
   // PF/AF/ZF/SF
   // Undefined
   {
@@ -354,7 +542,7 @@ void OpDispatchBuilder::GenerateFlags_MUL(FEXCore::X86Tables::DecodedOp Op, Orde
     // CF and OF are set if the result of the operation can't be fit in to the destination register
     // If the value can fit then the top bits will be zero
 
-    auto SignBit = _Sbfe(1, GetSrcSize(Op) * 8 - 1, Res);
+    auto SignBit = _Sbfe(1, SrcSize * 8 - 1, Res);
 
     auto SelectOp = _Select(FEXCore::IR::COND_EQ, High, SignBit, _Constant(0), _Constant(1));
 
@@ -363,7 +551,7 @@ void OpDispatchBuilder::GenerateFlags_MUL(FEXCore::X86Tables::DecodedOp Op, Orde
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_UMUL(FEXCore::X86Tables::DecodedOp Op, OrderedNode *High) {
+void OpDispatchBuilder::CalculcateFlags_UMUL(OrderedNode *High) {
   // AF/SF/PF/ZF
   // Undefined
   {
@@ -385,7 +573,7 @@ void OpDispatchBuilder::GenerateFlags_UMUL(FEXCore::X86Tables::DecodedOp Op, Ord
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_Logical(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+void OpDispatchBuilder::CalculcateFlags_Logical(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
   // AF
   {
     // Undefined
@@ -395,7 +583,7 @@ void OpDispatchBuilder::GenerateFlags_Logical(FEXCore::X86Tables::DecodedOp Op,
 
   // SF
   {
-    auto SignBitConst = _Constant(GetSrcSize(Op) * 8 - 1);
+    auto SignBitConst = _Constant(SrcSize * 8 - 1);
 
     auto LshrOp = _Lshr(Res, SignBitConst);
     SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(LshrOp);
@@ -430,11 +618,11 @@ auto oldflag = GetRFLAG(FEXCore::X86State::flag);\
 auto newval = _Select(FEXCore::IR::COND_EQ, cond, _Constant(0), oldflag, newflag);\
 SetRFLAG<FEXCore::X86State::flag>(newval);
 
-void OpDispatchBuilder::GenerateFlags_ShiftLeft(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+void OpDispatchBuilder::CalculcateFlags_ShiftLeft(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
   // CF
   {
     // Extract the last bit shifted in to CF
-    auto Size = _Constant(GetSrcSize(Op) * 8);
+    auto Size = _Constant(SrcSize * 8);
     auto ShiftAmt = _Sub(Size, Src2);
     auto LastBit = _And(_Lshr(Src1, ShiftAmt), _Constant(1));
     COND_FLAG_SET(Src2, RFLAG_CF_LOC, LastBit);
@@ -466,7 +654,7 @@ void OpDispatchBuilder::GenerateFlags_ShiftLeft(FEXCore::X86Tables::DecodedOp Op
 
   // SF
   {
-    auto val = _Bfe(1, GetSrcSize(Op) * 8 - 1, Res);
+    auto val = _Bfe(1, SrcSize * 8 - 1, Res);
     COND_FLAG_SET(Src2, RFLAG_SF_LOC, val);
   }
 
@@ -474,12 +662,12 @@ void OpDispatchBuilder::GenerateFlags_ShiftLeft(FEXCore::X86Tables::DecodedOp Op
   {
     // In the case of left shift. OF is only set from the result of <Top Source Bit> XOR <Top Result Bit>
     // When Shift > 1 then OF is undefined
-    auto val = _Bfe(1, GetSrcSize(Op) * 8 - 1, _Xor(Src1, Res));
+    auto val = _Bfe(1, SrcSize * 8 - 1, _Xor(Src1, Res));
     COND_FLAG_SET(Src2, RFLAG_OF_LOC, val);
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_ShiftRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+void OpDispatchBuilder::CalculcateFlags_ShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
   // CF
   {
     // Extract the last bit shifted in to CF
@@ -514,7 +702,7 @@ void OpDispatchBuilder::GenerateFlags_ShiftRight(FEXCore::X86Tables::DecodedOp O
 
   // SF
   {
-    auto val =_Bfe(1, GetSrcSize(Op) * 8 - 1, Res);
+    auto val =_Bfe(1, SrcSize * 8 - 1, Res);
     COND_FLAG_SET(Src2, RFLAG_SF_LOC, val);
   }
 
@@ -522,12 +710,12 @@ void OpDispatchBuilder::GenerateFlags_ShiftRight(FEXCore::X86Tables::DecodedOp O
   {
     // Only defined when Shift is 1 else undefined
     // OF flag is set if a sign change occurred
-    auto val = _Bfe(1, GetSrcSize(Op) * 8 - 1, _Xor(Src1, Res));
+    auto val = _Bfe(1, SrcSize * 8 - 1, _Xor(Src1, Res));
     COND_FLAG_SET(Src2, RFLAG_OF_LOC, val);
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_SignShiftRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+void OpDispatchBuilder::CalculcateFlags_SignShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
   // CF
   {
     // Extract the last bit shifted in to CF
@@ -562,7 +750,7 @@ void OpDispatchBuilder::GenerateFlags_SignShiftRight(FEXCore::X86Tables::Decoded
 
   // SF
   {
-    auto SignBitConst = _Constant(GetSrcSize(Op) * 8 - 1);
+    auto SignBitConst = _Constant(SrcSize * 8 - 1);
 
     auto LshrOp = _Lshr(Res, SignBitConst);
     COND_FLAG_SET(Src2, RFLAG_SF_LOC, LshrOp);
@@ -574,14 +762,14 @@ void OpDispatchBuilder::GenerateFlags_SignShiftRight(FEXCore::X86Tables::Decoded
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_ShiftLeftImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+void OpDispatchBuilder::CalculcateFlags_ShiftLeftImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
   // No flags changed if shift is zero
   if (Shift == 0) return;
 
   // CF
   {
     // Extract the last bit shifted in to CF
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(_Bfe(1, GetSrcSize(Op) * 8 - Shift, Src1));
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(_Bfe(1, SrcSize * 8 - Shift, Src1));
   }
 
   // PF
@@ -610,20 +798,20 @@ void OpDispatchBuilder::GenerateFlags_ShiftLeftImmediate(FEXCore::X86Tables::Dec
 
   // SF
   {
-    auto LshrOp = _Bfe(1, GetSrcSize(Op) * 8 - 1, Res);
+    auto LshrOp = _Bfe(1, SrcSize * 8 - 1, Res);
 
     SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(LshrOp);
 
     // OF
     // In the case of left shift. OF is only set from the result of <Top Source Bit> XOR <Top Result Bit>
     if (Shift == 1) {
-      auto SourceBit = _Bfe(1, GetSrcSize(Op) * 8 - 1, Src1);
+      auto SourceBit = _Bfe(1, SrcSize * 8 - 1, Src1);
       SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(_Xor(SourceBit, LshrOp));
     }
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_SignShiftRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+void OpDispatchBuilder::CalculcateFlags_SignShiftRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
   // No flags changed if shift is zero
   if (Shift == 0) return;
 
@@ -659,7 +847,7 @@ void OpDispatchBuilder::GenerateFlags_SignShiftRightImmediate(FEXCore::X86Tables
 
   // SF
   {
-    auto SignBitConst = _Constant(GetSrcSize(Op) * 8 - 1);
+    auto SignBitConst = _Constant(SrcSize * 8 - 1);
 
     auto LshrOp = _Lshr(Res, SignBitConst);
     SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(LshrOp);
@@ -674,7 +862,7 @@ void OpDispatchBuilder::GenerateFlags_SignShiftRightImmediate(FEXCore::X86Tables
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_ShiftRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+void OpDispatchBuilder::CalculcateFlags_ShiftRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
   // No flags changed if shift is zero
   if (Shift == 0) return;
 
@@ -710,7 +898,7 @@ void OpDispatchBuilder::GenerateFlags_ShiftRightImmediate(FEXCore::X86Tables::De
 
   // SF
   {
-    auto SignBitConst = _Constant(GetSrcSize(Op) * 8 - 1);
+    auto SignBitConst = _Constant(SrcSize * 8 - 1);
 
     auto LshrOp = _Lshr(Res, SignBitConst);
     SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(LshrOp);
@@ -721,13 +909,13 @@ void OpDispatchBuilder::GenerateFlags_ShiftRightImmediate(FEXCore::X86Tables::De
     // Only defined when Shift is 1 else undefined
     // Is set to the MSB of the original value
     if (Shift == 1) {
-      SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(_Bfe(1, GetSrcSize(Op) * 8 - 1, Src1));
+      SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(_Bfe(1, SrcSize * 8 - 1, Src1));
     }
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_RotateRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-  auto OpSize = GetSrcSize(Op) * 8;
+void OpDispatchBuilder::CalculcateFlags_RotateRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+  auto OpSize = SrcSize * 8;
 
   // Extract the last bit shifted in to CF
   auto NewCF = _Bfe(1, OpSize - 1, Res);
@@ -755,8 +943,8 @@ void OpDispatchBuilder::GenerateFlags_RotateRight(FEXCore::X86Tables::DecodedOp
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_RotateLeft(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-  auto OpSize = GetSrcSize(Op) * 8;
+void OpDispatchBuilder::CalculcateFlags_RotateLeft(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+  auto OpSize = SrcSize * 8;
 
   // Extract the last bit shifted in to CF
   //auto Size = _Constant(GetSrcSize(Res) * 8);
@@ -785,10 +973,10 @@ void OpDispatchBuilder::GenerateFlags_RotateLeft(FEXCore::X86Tables::DecodedOp O
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_RotateRightImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+void OpDispatchBuilder::CalculcateFlags_RotateRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
   if (Shift == 0) return;
 
-  auto OpSize = GetSrcSize(Op) * 8;
+  auto OpSize = SrcSize * 8;
 
   auto NewCF = _Bfe(1, OpSize - Shift, Src1);
 
@@ -807,10 +995,10 @@ void OpDispatchBuilder::GenerateFlags_RotateRightImmediate(FEXCore::X86Tables::D
   }
 }
 
-void OpDispatchBuilder::GenerateFlags_RotateLeftImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+void OpDispatchBuilder::CalculcateFlags_RotateLeftImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
   if (Shift == 0) return;
 
-  auto OpSize = GetSrcSize(Op) * 8;
+  auto OpSize = SrcSize * 8;
 
   // CF
   {
@@ -827,4 +1015,264 @@ void OpDispatchBuilder::GenerateFlags_RotateLeftImmediate(FEXCore::X86Tables::De
   }
 }
 
+void OpDispatchBuilder::CalculcateFlags_FCMP(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
+  OrderedNode *HostFlag_CF = _GetHostFlag(Res, FCMP_FLAG_LT);
+  OrderedNode *HostFlag_ZF = _GetHostFlag(Res, FCMP_FLAG_EQ);
+  OrderedNode *HostFlag_Unordered  = _GetHostFlag(Res, FCMP_FLAG_UNORDERED);
+
+  SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(HostFlag_CF);
+  SetRFLAG<FEXCore::X86State::RFLAG_ZF_LOC>(HostFlag_ZF);
+  SetRFLAG<FEXCore::X86State::RFLAG_PF_LOC>(HostFlag_Unordered);
+
+  auto ZeroConst = _Constant(0);
+  SetRFLAG<FEXCore::X86State::RFLAG_AF_LOC>(ZeroConst);
+  SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(ZeroConst);
+  SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(ZeroConst);
+}
+
+void OpDispatchBuilder::CalculcateFlags_BEXTR(OrderedNode *Src) {
+  // Handle flag setting.
+  //
+  // All that matters primarily for this instruction is
+  // that we only set the ZF flag properly.
+  //
+  // CF and OF are defined as being set to zero
+  //
+  SetRFLAG<X86State::RFLAG_CF_LOC>(_Constant(0));
+  SetRFLAG<X86State::RFLAG_OF_LOC>(_Constant(0));
+
+  // Every other flag is considered undefined after a
+  // BEXTR instruction, but we opt to reliably clear them.
+  //
+  SetRFLAG<X86State::RFLAG_AF_LOC>(_Constant(0));
+  SetRFLAG<X86State::RFLAG_SF_LOC>(_Constant(0));
+
+  // PF
+  if (CTX->Config.ABINoPF) {
+    _InvalidateFlags(1UL << X86State::RFLAG_PF_LOC);
+  } else {
+    SetRFLAG<X86State::RFLAG_PF_LOC>(_Constant(0));
+  }
+
+  // ZF
+  auto ZeroOp = _Select(IR::COND_EQ,
+                        Src, _Constant(0),
+                        _Constant(1), _Constant(0));
+  SetRFLAG<X86State::RFLAG_ZF_LOC>(ZeroOp);
+}
+
+void OpDispatchBuilder::CalculcateFlags_BLSI(uint8_t SrcSize, OrderedNode *Src) {
+  // Now for the flags:
+  //
+  // Only CF, SF, ZF and OF are defined as being updated
+  // CF is cleared if Src is zero, otherwise it's set.
+  // SF is set to the value of the most significant operand bit of Result.
+  // OF is always cleared
+  // ZF is set, as usual, if Result is zero or not.
+  //
+  // AF and PF are documented as being in an undefined state after
+  // a BLSI operation, however, we choose to reliably clear them.
+
+  auto Zero = _Constant(0);
+  auto One = _Constant(1);
+
+  SetRFLAG<X86State::RFLAG_OF_LOC>(Zero);
+  SetRFLAG<X86State::RFLAG_AF_LOC>(Zero);
+  if (CTX->Config.ABINoPF) {
+    _InvalidateFlags(1UL << X86State::RFLAG_PF_LOC);
+  } else {
+    SetRFLAG<X86State::RFLAG_PF_LOC>(Zero);
+  }
+
+  // ZF
+  {
+    auto ZFOp = _Select(IR::COND_EQ,
+                        Src, Zero,
+                        One, Zero);
+    SetRFLAG<X86State::RFLAG_ZF_LOC>(ZFOp);
+  }
+
+  // CF
+  {
+    auto CFOp = _Select(IR::COND_EQ,
+                        Src, Zero,
+                        Zero, One);
+    SetRFLAG<X86State::RFLAG_CF_LOC>(CFOp);
+  }
+
+  // SF
+  {
+    auto SignBit = _Constant(SrcSize * 8 - 1);
+    auto SFOp = _Lshr(Src, SignBit);
+
+    SetRFLAG<X86State::RFLAG_SF_LOC>(SFOp);
+  }
+}
+
+void OpDispatchBuilder::CalculcateFlags_BLSMSK(OrderedNode *Src) {
+  // Now for the flags.
+
+  auto Zero = _Constant(0);
+  auto One = _Constant(1);
+
+  SetRFLAG<X86State::RFLAG_ZF_LOC>(Zero);
+  SetRFLAG<X86State::RFLAG_OF_LOC>(Zero);
+  SetRFLAG<X86State::RFLAG_AF_LOC>(Zero);
+  if (CTX->Config.ABINoPF) {
+    _InvalidateFlags(1UL << X86State::RFLAG_PF_LOC);
+  } else {
+    SetRFLAG<X86State::RFLAG_PF_LOC>(Zero);
+  }
+
+  auto CFOp = _Select(IR::COND_EQ,
+                      Src, Zero,
+                      Zero, One);
+  SetRFLAG<X86State::RFLAG_CF_LOC>(CFOp);
+}
+
+void OpDispatchBuilder::CalculcateFlags_BLSR(uint8_t SrcSize, OrderedNode *Result, OrderedNode *Src) {
+  // Now for flags.
+  auto Zero = _Constant(0);
+  auto One = _Constant(1);
+
+  SetRFLAG<X86State::RFLAG_OF_LOC>(Zero);
+  SetRFLAG<X86State::RFLAG_AF_LOC>(Zero);
+  if (CTX->Config.ABINoPF) {
+    _InvalidateFlags(1UL << X86State::RFLAG_PF_LOC);
+  } else {
+    SetRFLAG<X86State::RFLAG_PF_LOC>(Zero);
+  }
+
+  // ZF
+  {
+    auto ZFOp = _Select(IR::COND_EQ,
+                        Result, Zero,
+                        One, Zero);
+    SetRFLAG<X86State::RFLAG_ZF_LOC>(ZFOp);
+  }
+
+  // CF
+  {
+    auto CFOp = _Select(IR::COND_EQ,
+                        Src, Zero,
+                        Zero, One);
+    SetRFLAG<X86State::RFLAG_CF_LOC>(CFOp);
+  }
+
+  // SF
+  {
+    auto SignBit = _Constant(SrcSize * 8 - 1);
+    auto SFOp = _Lshr(Result, SignBit);
+
+    SetRFLAG<X86State::RFLAG_SF_LOC>(SFOp);
+  }
+}
+
+void OpDispatchBuilder::CalculcateFlags_POPCOUNT(OrderedNode *Src) {
+  // Set ZF
+  auto Zero = _Constant(0);
+  auto ZFResult = _Select(FEXCore::IR::COND_EQ,
+      Src,  Zero,
+      _Constant(1), Zero);
+
+  // Set flags
+  SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(Zero);
+  SetRFLAG<FEXCore::X86State::RFLAG_PF_LOC>(Zero);
+  SetRFLAG<FEXCore::X86State::RFLAG_AF_LOC>(Zero);
+  SetRFLAG<FEXCore::X86State::RFLAG_ZF_LOC>(ZFResult);
+  SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(Zero);
+  SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(Zero);
+}
+
+void OpDispatchBuilder::CalculcateFlags_BZHI(uint8_t SrcSize, OrderedNode *Result, OrderedNode *Src) {
+  // Now for the flags
+
+  auto Bounds = _Constant(SrcSize * 8- 1);
+  auto Zero = _Constant(0);
+  auto One = _Constant(1);
+
+  SetRFLAG<X86State::RFLAG_OF_LOC>(Zero);
+  SetRFLAG<X86State::RFLAG_AF_LOC>(Zero);
+  if (CTX->Config.ABINoPF) {
+    _InvalidateFlags(1UL << X86State::RFLAG_PF_LOC);
+  } else {
+    SetRFLAG<X86State::RFLAG_PF_LOC>(Zero);
+  }
+
+  // ZF
+  {
+    auto ZFOp = _Select(IR::COND_EQ,
+                        Result, Zero,
+                        One, Zero);
+    SetRFLAG<X86State::RFLAG_ZF_LOC>(ZFOp);
+  }
+
+  // CF
+  {
+    auto CFOp = _Select(IR::COND_UGT,
+                        Src, Bounds,
+                        One, Zero);
+    SetRFLAG<X86State::RFLAG_CF_LOC>(CFOp);
+  }
+
+  // SF
+  {
+    auto SFOp = _Lshr(Result, Bounds);
+
+    SetRFLAG<X86State::RFLAG_SF_LOC>(SFOp);
+  }
+}
+
+void OpDispatchBuilder::CalculcateFlags_TZCNT(OrderedNode *Src) {
+  // OF, SF, AF, PF all undefined
+  auto Zero = _Constant(0);
+  auto ZFResult = _Select(FEXCore::IR::COND_EQ,
+      Src,  Zero,
+      _Constant(1), Zero);
+
+  // Set flags
+  SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(ZFResult);
+  SetRFLAG<FEXCore::X86State::RFLAG_ZF_LOC>(_Bfe(1, 0, Src));
+}
+
+void OpDispatchBuilder::CalculcateFlags_LZCNT(uint8_t SrcSize, OrderedNode *Src) {
+  // OF, SF, AF, PF all undefined
+
+  auto Zero = _Constant(0);
+  auto ZFResult = _Select(FEXCore::IR::COND_EQ,
+      Src,  Zero,
+      _Constant(1), Zero);
+
+  // Set flags
+  SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(ZFResult);
+  SetRFLAG<FEXCore::X86State::RFLAG_ZF_LOC>(_Bfe(1, SrcSize * 8 - 1, Src));
+}
+
+void OpDispatchBuilder::CalculcateFlags_BITSELECT(OrderedNode *Src) {
+  // OF, SF, AF, PF, CF all undefined
+
+  auto ZeroConst = _Constant(0);
+  auto OneConst = _Constant(1);
+
+  // ZF is set to 1 if the source was zero
+  auto ZFSelectOp = _Select(FEXCore::IR::COND_EQ,
+      Src, ZeroConst,
+      OneConst, ZeroConst);
+
+  SetRFLAG<FEXCore::X86State::RFLAG_ZF_LOC>(ZFSelectOp);
+}
+
+void OpDispatchBuilder::CalculcateFlags_RDRAND(OrderedNode *Src) {
+  // OF, SF, ZF, AF, PF all zero
+  // CF is set to the incoming source
+
+  auto ZeroConst = _Constant(0);
+  SetRFLAG<X86State::RFLAG_OF_LOC>(ZeroConst);
+  SetRFLAG<X86State::RFLAG_SF_LOC>(ZeroConst);
+  SetRFLAG<X86State::RFLAG_ZF_LOC>(ZeroConst);
+  SetRFLAG<X86State::RFLAG_AF_LOC>(ZeroConst);
+  SetRFLAG<X86State::RFLAG_PF_LOC>(ZeroConst);
+  SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(Src);
+}
+
 }

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

@@ -241,6 +241,8 @@ void OpDispatchBuilder::VectorALUOp<IR::OP_VCMPGT, 2>(OpcodeArgs);
 template
 void OpDispatchBuilder::VectorALUOp<IR::OP_VCMPGT, 4>(OpcodeArgs);
 template
+void OpDispatchBuilder::VectorALUOp<IR::OP_VCMPGT, 8>(OpcodeArgs);
+template
 void OpDispatchBuilder::VectorALUOp<IR::OP_VCMPEQ, 1>(OpcodeArgs);
 template
 void OpDispatchBuilder::VectorALUOp<IR::OP_VCMPEQ, 2>(OpcodeArgs);
@@ -295,6 +297,24 @@ void OpDispatchBuilder::VectorALUOp<IR::OP_VUQSUB, 1>(OpcodeArgs);
 template
 void OpDispatchBuilder::VectorALUOp<IR::OP_VUQSUB, 2>(OpcodeArgs);
 
+template<FEXCore::IR::IROps IROp, size_t ElementSize>
+void OpDispatchBuilder::VectorALUROp(OpcodeArgs) {
+  auto Size = GetSrcSize(Op);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+
+  auto ALUOp = _VAdd(Size, ElementSize, Src, Dest);
+  // Overwrite our IR's op type
+  ALUOp.first->Header.Op = IROp;
+
+  StoreResult(FPRClass, Op, ALUOp, -1);
+}
+
+template
+void OpDispatchBuilder::VectorALUROp<IR::OP_VFSUB, 4>(OpcodeArgs);
+template
+void OpDispatchBuilder::VectorALUROp<IR::OP_VFSUB, 8>(OpcodeArgs);
+
 template<FEXCore::IR::IROps IROp, size_t ElementSize>
 void OpDispatchBuilder::VectorScalarALUOp(OpcodeArgs) {
   auto Size = GetSrcSize(Op);
@@ -390,14 +410,35 @@ void OpDispatchBuilder::VectorUnaryOp<IR::OP_VABS, 2, false>(OpcodeArgs);
 template
 void OpDispatchBuilder::VectorUnaryOp<IR::OP_VABS, 4, false>(OpcodeArgs);
 
+template<FEXCore::IR::IROps IROp, size_t ElementSize>
+void OpDispatchBuilder::VectorUnaryDuplicateOp(OpcodeArgs) {
+  auto Size = GetSrcSize(Op);
+
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  auto ALUOp = _VFSqrt(ElementSize, ElementSize, Src);
+  // Overwrite our IR's op type
+  ALUOp.first->Header.Op = IROp;
+
+  // Duplicate the lower bits
+  auto Result = _VDupElement(Size, ElementSize, ALUOp, 0);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+template
+void OpDispatchBuilder::VectorUnaryDuplicateOp<IR::OP_VFRSQRT, 4>(OpcodeArgs);
+template
+void OpDispatchBuilder::VectorUnaryDuplicateOp<IR::OP_VFRECP, 4>(OpcodeArgs);
+
 void OpDispatchBuilder::MOVQOp(OpcodeArgs) {
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
   // This instruction is a bit special that if the destination is a register then it'll ZEXT the 64bit source to 128bit
   if (Op->Dest.IsGPR()) {
     const auto gpr = Op->Dest.Data.GPR.GPR;
-    _StoreContext(FPRClass, 8, offsetof(FEXCore::Core::CPUState, xmm[gpr - FEXCore::X86State::REG_XMM_0][0]), Src);
+
+    _StoreContext(8, FPRClass, Src, offsetof(FEXCore::Core::CPUState, xmm[gpr - FEXCore::X86State::REG_XMM_0][0]));
     auto Const = _Constant(0);
-    _StoreContext(GPRClass, 8, offsetof(FEXCore::Core::CPUState, xmm[gpr - FEXCore::X86State::REG_XMM_0][1]), Const);
+    _StoreContext(8, GPRClass, Const, offsetof(FEXCore::Core::CPUState, xmm[gpr - FEXCore::X86State::REG_XMM_0][1]));
   }
   else {
     // This is simple, just store the result
@@ -438,14 +479,15 @@ void OpDispatchBuilder::MOVMSKOpOne(OpcodeArgs) {
   //TODO: We could remove this VCastFromGOR + VInsGPR pair if we had a VDUPFromGPR instruction that maps directly to AArch64.
   auto M = _Constant(0x80'40'20'10'08'04'02'01ULL);
   OrderedNode *VMask = _VCastFromGPR(16, 8, M);
-  VMask = _VInsGPR(16, 8, VMask, M, 1);
 
-  auto VCMP = _VCMPLTZ(Src, 16, 1);
-  auto VAnd = _VAnd(VCMP, VMask, 16, 1);
+  VMask = _VInsGPR(16, 8, 1, VMask, M);
 
-  auto VAdd1 = _VAddP(VAnd, VAnd, 16, 1);
-  auto VAdd2 = _VAddP(VAdd1, VAdd1, 8, 1);
-  auto VAdd3 = _VAddP(VAdd2, VAdd2, 8, 1);
+  auto VCMP = _VCMPLTZ(16, 1, Src);
+  auto VAnd = _VAnd(16, 1, VCMP, VMask);
+
+  auto VAdd1 = _VAddP(16, 1, VAnd, VAnd);
+  auto VAdd2 = _VAddP(8, 1, VAdd1, VAdd1);
+  auto VAdd3 = _VAddP(8, 1, VAdd2, VAdd2);
 
   StoreResult(GPRClass, Op, _VExtractToGPR(16, 2, VAdd3, 0), -1);
 }
@@ -502,11 +544,11 @@ void OpDispatchBuilder::PSHUFBOp(OpcodeArgs) {
   // Bits [6:4] is reserved for 128bit
   // Bits [6:3] is reserved for 64bit
   if (Size == 8) {
-    auto MaskVector = _VectorImm(0b1000'0111, Size, 1);
+    auto MaskVector = _VectorImm(Size, 1, 0b1000'0111);
     Src = _VAnd(Size, Size, Src, MaskVector);
   }
   else {
-    auto MaskVector = _VectorImm(0b1000'1111, Size, 1);
+    auto MaskVector = _VectorImm(Size, 1, 0b1000'1111);
     Src = _VAnd(Size, Size, Src, MaskVector);
   }
   auto Res = _VTBL1(Size, Dest, Src);
@@ -627,7 +669,7 @@ void OpDispatchBuilder::PINSROp(OpcodeArgs) {
   Index &= NumElements - 1;
 
   // This maps 1:1 to an AArch64 NEON Op
-  auto ALUOp = _VInsGPR(Size, ElementSize, Dest, Src, Index);
+  auto ALUOp = _VInsGPR(Size, ElementSize, Index, Dest, Src);
   StoreResult(FPRClass, Op, ALUOp, -1);
 }
 
@@ -670,10 +712,10 @@ void OpDispatchBuilder::InsertPSOp(OpcodeArgs) {
 
   // ZMask happens after insert
   if (ZMask == 0xF) {
-    Dest = _VectorImm(0, 16, 4);
+    Dest = _VectorImm(16, 4, 0);
   }
   else if (ZMask) {
-    auto Zero = _VectorImm(0, 16, 4);
+    auto Zero = _VectorImm(16, 4, 0);
     for (size_t i = 0; i < 4; ++i) {
       if (ZMask & (1 << i)) {
         Dest = _VInsElement(GetDstSize(Op), 4, i, 0, Dest, Zero);
@@ -766,7 +808,7 @@ void OpDispatchBuilder::PSRLDOp(OpcodeArgs) {
   OrderedNode *Result{};
 
   // Incoming element size for the shift source is always 8
-  auto MaxShift = _VectorImm(ElementSize * 8, 8, 8);
+  auto MaxShift = _VectorImm(8, 8, ElementSize * 8);
   Src = _VUMin(8, 8, MaxShift, Src);
   Result = _VUShrS(Size, ElementSize, Dest, Src);
 
@@ -830,7 +872,7 @@ void OpDispatchBuilder::PSLL(OpcodeArgs) {
   OrderedNode *Result{};
 
   // Incoming element size for the shift source is always 8
-  auto MaxShift = _VectorImm(ElementSize * 8, 8, 8);
+  auto MaxShift = _VectorImm(8, 8, ElementSize * 8);
   Src = _VUMin(8, 8, MaxShift, Src);
   Result = _VUShlS(Size, ElementSize, Dest, Src);
 
@@ -854,7 +896,7 @@ void OpDispatchBuilder::PSRAOp(OpcodeArgs) {
   OrderedNode *Result{};
 
   // Incoming element size for the shift source is always 8
-  auto MaxShift = _VectorImm(ElementSize * 8, 8, 8);
+  auto MaxShift = _VectorImm(8, 8, ElementSize * 8);
   Src = _VUMin(8, 8, MaxShift, Src);
   Result = _VSShrS(Size, ElementSize, Dest, Src);
 
@@ -957,12 +999,11 @@ void OpDispatchBuilder::CVTFPR_To_GPR(OpcodeArgs) {
   // Source Element size is determined by instruction
   size_t GPRSize = GetDstSize(Op);
 
-  size_t ElementSize = SrcElementSize;
   if constexpr (HostRoundingMode) {
-    Src = _Float_ToGPR_S(Src, ElementSize, GPRSize);
+    Src = _Float_ToGPR_S(GPRSize, SrcElementSize, Src);
   }
   else {
-    Src = _Float_ToGPR_ZS(Src, ElementSize, GPRSize);
+    Src = _Float_ToGPR_ZS(GPRSize, SrcElementSize, Src);
   }
 
   StoreResult_WithOpSize(GPRClass, Op, Op->Dest, Src, GPRSize, -1);
@@ -985,11 +1026,11 @@ void OpDispatchBuilder::Vector_CVT_Int_To_Float(OpcodeArgs) {
   size_t ElementSize = SrcElementSize;
   size_t Size = GetDstSize(Op);
   if constexpr (Widen) {
-    Src = _VSXTL(Src, Size, ElementSize);
+    Src = _VSXTL(Size, ElementSize, Src);
     ElementSize <<= 1;
   }
 
-  Src = _Vector_SToF(Src, Size, ElementSize);
+  Src = _Vector_SToF(Size, ElementSize, Src);
 
   StoreResult(FPRClass, Op, Src, -1);
 }
@@ -1007,15 +1048,15 @@ void OpDispatchBuilder::Vector_CVT_Float_To_Int(OpcodeArgs) {
   size_t Size = GetDstSize(Op);
 
   if constexpr (Narrow) {
-    Src = _Vector_FToF(Size, SrcElementSize >> 1, SrcElementSize, Src);
+    Src = _Vector_FToF(Size, SrcElementSize >> 1, Src, SrcElementSize);
     ElementSize >>= 1;
   }
 
   if constexpr (HostRoundingMode) {
-    Src = _Vector_FToS(Src, Size, ElementSize);
+    Src = _Vector_FToS(Size, ElementSize, Src);
   }
   else {
-    Src = _Vector_FToZS(Src, Size, ElementSize);
+    Src = _Vector_FToZS(Size, ElementSize, Src);
   }
 
   StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src, Size, -1);
@@ -1025,6 +1066,8 @@ template
 void OpDispatchBuilder::Vector_CVT_Float_To_Int<4, false, false>(OpcodeArgs);
 template
 void OpDispatchBuilder::Vector_CVT_Float_To_Int<4, false, true>(OpcodeArgs);
+template
+void OpDispatchBuilder::Vector_CVT_Float_To_Int<4, true, false>(OpcodeArgs);
 
 template
 void OpDispatchBuilder::Vector_CVT_Float_To_Int<8, true, true>(OpcodeArgs);
@@ -1053,10 +1096,10 @@ void OpDispatchBuilder::Vector_CVT_Float_To_Float(OpcodeArgs) {
   size_t Size = GetDstSize(Op);
 
   if constexpr (DstElementSize > SrcElementSize) {
-    Src = _Vector_FToF(Size, SrcElementSize << 1, SrcElementSize, Src);
+    Src = _Vector_FToF(Size, SrcElementSize << 1, Src, SrcElementSize);
   }
   else {
-    Src = _Vector_FToF(Size, SrcElementSize >> 1, SrcElementSize, Src);
+    Src = _Vector_FToF(Size, SrcElementSize >> 1, Src, SrcElementSize);
   }
 
   StoreResult(FPRClass, Op, Src, -1);
@@ -1074,12 +1117,12 @@ void OpDispatchBuilder::MMX_To_XMM_Vector_CVT_Int_To_Float(OpcodeArgs) {
   size_t ElementSize = SrcElementSize;
   size_t DstSize = GetDstSize(Op);
   if constexpr (Widen) {
-    Src = _VSXTL(Src, DstSize, ElementSize);
+    Src = _VSXTL(DstSize, ElementSize, Src);
     ElementSize <<= 1;
   }
 
   // Always signed
-  Src = _Vector_SToF(Src, DstSize, ElementSize);
+  Src = _Vector_SToF(DstSize, ElementSize, Src);
 
   OrderedNode *Dest{};
   if constexpr (Widen) {
@@ -1107,14 +1150,14 @@ void OpDispatchBuilder::XMM_To_MMX_Vector_CVT_Float_To_Int(OpcodeArgs) {
   size_t Size = GetDstSize(Op);
 
   // Always narrows
-  Src = _Vector_FToF(Size, SrcElementSize >> 1, SrcElementSize, Src);
+  Src = _Vector_FToF(Size, SrcElementSize >> 1, Src, SrcElementSize);
   ElementSize >>= 1;
 
   if constexpr (HostRoundingMode) {
-    Src = _Vector_FToS(Src, Size, ElementSize);
+    Src = _Vector_FToS(Size, ElementSize, Src);
   }
   else {
-    Src = _Vector_FToZS(Src, Size, ElementSize);
+    Src = _Vector_FToZS(Size, ElementSize, Src);
   }
 
   StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src, Size, -1);
@@ -1133,7 +1176,7 @@ void OpDispatchBuilder::MASKMOVOp(OpcodeArgs) {
   OrderedNode *Src = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, -1);
   OrderedNode *Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, -1);
 
-  OrderedNode *MemDest = _LoadContext(GPRSize, GPROffset(X86State::REG_RDI), GPRClass);
+  OrderedNode *MemDest = _LoadContext(GPRSize, GPRClass, GPROffset(X86State::REG_RDI));
 
   const size_t NumElements = Size / 64;
   for (size_t Element = 0; Element < NumElements; ++Element) {
@@ -1187,13 +1230,6 @@ void OpDispatchBuilder::VFCMPOp(OpcodeArgs) {
   auto Size = GetSrcSize(Op);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
   OrderedNode *Dest = LoadSource_WithOpSize(FPRClass, Op, Op->Dest, GetDstSize(Op), Op->Flags, -1);
-  OrderedNode *Src2{};
-  if constexpr (Scalar) {
-    Src2 = _VExtractElement(GetDstSize(Op), Size, Dest, 0);
-  }
-  else {
-    Src2 = Dest;
-  }
   uint8_t CompType = Op->Src[1].Data.Literal.Value;
 
   OrderedNode *Result{};
@@ -1201,30 +1237,30 @@ void OpDispatchBuilder::VFCMPOp(OpcodeArgs) {
   //auto ALUOp = _VCMPGT(Size, ElementSize, Dest, Src);
   switch (CompType) {
     case 0x00: case 0x08: case 0x10: case 0x18: // EQ
-      Result = _VFCMPEQ(Size, ElementSize, Src2, Src);
+      Result = _VFCMPEQ(Size, ElementSize, Dest, Src);
     break;
     case 0x01: case 0x09: case 0x11: case 0x19: // LT, GT(Swapped operand)
-      Result = _VFCMPLT(Size, ElementSize, Src2, Src);
+      Result = _VFCMPLT(Size, ElementSize, Dest, Src);
     break;
     case 0x02: case 0x0A: case 0x12: case 0x1A: // LE, GE(Swapped operand)
-      Result = _VFCMPLE(Size, ElementSize, Src2, Src);
+      Result = _VFCMPLE(Size, ElementSize, Dest, Src);
     break;
     case 0x03: case 0x0B: case 0x13: case 0x1B: // Unordered
-      Result = _VFCMPUNO(Size, ElementSize, Src2, Src);
+      Result = _VFCMPUNO(Size, ElementSize, Dest, Src);
     break;
     case 0x04: case 0x0C: case 0x14: case 0x1C: // NEQ
-      Result = _VFCMPNEQ(Size, ElementSize, Src2, Src);
+      Result = _VFCMPNEQ(Size, ElementSize, Dest, Src);
     break;
     case 0x05: case 0x0D: case 0x15: case 0x1D: // NLT, NGT(Swapped operand)
-      Result = _VFCMPLT(Size, ElementSize, Src2, Src);
+      Result = _VFCMPLT(Size, ElementSize, Dest, Src);
       Result = _VNot(Size, ElementSize, Result);
     break;
     case 0x06: case 0x0E: case 0x16: case 0x1E: // NLE, NGE(Swapped operand)
-      Result = _VFCMPLE(Size, ElementSize, Src2, Src);
+      Result = _VFCMPLE(Size, ElementSize, Dest, Src);
       Result = _VNot(Size, ElementSize, Result);
     break;
     case 0x07: case 0x0F: case 0x17: case 0x1F: // Ordered
-      Result = _VFCMPORD(Size, ElementSize, Src2, Src);
+      Result = _VFCMPORD(Size, ElementSize, Dest, Src);
     break;
     default:
       LOGMAN_MSG_A_FMT("Unknown Comparison type: {}", CompType);
@@ -1268,7 +1304,7 @@ void OpDispatchBuilder::FXSaveOp(OpcodeArgs) {
   }
 
   {
-    auto FCW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FCW), GPRClass);
+    auto FCW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
     _StoreMem(GPRClass, 2, Mem, FCW, 2);
   }
 
@@ -1294,7 +1330,7 @@ void OpDispatchBuilder::FXSaveOp(OpcodeArgs) {
   {
     // FTW
     OrderedNode *MemLocation = _Add(Mem, _Constant(4));
-    auto FTW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FTW), GPRClass);
+    auto FTW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FTW));
     _StoreMem(GPRClass, 2, MemLocation, FTW, 2);
   }
 
@@ -1343,7 +1379,7 @@ void OpDispatchBuilder::FXSaveOp(OpcodeArgs) {
   // If OSFXSR bit in CR4 is not set than FXSAVE /may/ not save the XMM registers
   // This is implementation dependent
   for (unsigned i = 0; i < 8; ++i) {
-    OrderedNode *MMReg = _LoadContext(16, offsetof(FEXCore::Core::CPUState, mm[i]), FPRClass);
+    OrderedNode *MMReg = _LoadContext(16, FPRClass, offsetof(FEXCore::Core::CPUState, mm[i]));
     OrderedNode *MemLocation = _Add(Mem, _Constant(i * 16 + 32));
 
     _StoreMem(FPRClass, 16, MemLocation, MMReg, 16);
@@ -1351,7 +1387,7 @@ void OpDispatchBuilder::FXSaveOp(OpcodeArgs) {
   unsigned NumRegs = CTX->Config.Is64BitMode ? 16 : 8;
 
   for (unsigned i = 0; i < NumRegs; ++i) {
-    OrderedNode *XMMReg = _LoadContext(16, offsetof(FEXCore::Core::CPUState, xmm[i]), FPRClass);
+    OrderedNode *XMMReg = _LoadContext(16, FPRClass, offsetof(FEXCore::Core::CPUState, xmm[i]));
     OrderedNode *MemLocation = _Add(Mem, _Constant(i * 16 + 160));
 
     _StoreMem(FPRClass, 16, MemLocation, XMMReg, 16);
@@ -1364,7 +1400,7 @@ void OpDispatchBuilder::FXRStoreOp(OpcodeArgs) {
 
   auto NewFCW = _LoadMem(GPRClass, 2, Mem, 2);
   _F80LoadFCW(NewFCW);
-  _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FCW), NewFCW);
+  _StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 
   {
     OrderedNode *MemLocation = _Add(Mem, _Constant(2));
@@ -1389,20 +1425,20 @@ void OpDispatchBuilder::FXRStoreOp(OpcodeArgs) {
     // FTW
     OrderedNode *MemLocation = _Add(Mem, _Constant(4));
     auto NewFTW = _LoadMem(GPRClass, 2, MemLocation, 2);
-    _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FTW), NewFTW);
+    _StoreContext(2, GPRClass, NewFTW, offsetof(FEXCore::Core::CPUState, FTW));
   }
 
   for (unsigned i = 0; i < 8; ++i) {
     OrderedNode *MemLocation = _Add(Mem, _Constant(i * 16 + 32));
     auto MMReg = _LoadMem(FPRClass, 16, MemLocation, 16);
-    _StoreContext(FPRClass, 16, offsetof(FEXCore::Core::CPUState, mm[i]), MMReg);
+    _StoreContext(16, FPRClass, MMReg, offsetof(FEXCore::Core::CPUState, mm[i]));
   }
   unsigned NumRegs = CTX->Config.Is64BitMode ? 16 : 8;
 
   for (unsigned i = 0; i < NumRegs; ++i) {
     OrderedNode *MemLocation = _Add(Mem, _Constant(i * 16 + 160));
     auto XMMReg = _LoadMem(FPRClass, 16, MemLocation, 16);
-    _StoreContext(FPRClass, 16, offsetof(FEXCore::Core::CPUState, xmm[i]), XMMReg);
+    _StoreContext(16, FPRClass, XMMReg, offsetof(FEXCore::Core::CPUState, xmm[i]));
   }
 }
 
@@ -1427,23 +1463,12 @@ template<size_t ElementSize>
 void OpDispatchBuilder::UCOMISxOp(OpcodeArgs) {
   OrderedNode *Src1 = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   OrderedNode *Src2 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
-  OrderedNode *Res = _FCmp(Src1, Src2, ElementSize,
+  OrderedNode *Res = _FCmp(ElementSize, Src1, Src2,
     (1 << FCMP_FLAG_EQ) |
     (1 << FCMP_FLAG_LT) |
     (1 << FCMP_FLAG_UNORDERED));
 
-  OrderedNode *HostFlag_CF = _GetHostFlag(Res, FCMP_FLAG_LT);
-  OrderedNode *HostFlag_ZF = _GetHostFlag(Res, FCMP_FLAG_EQ);
-  OrderedNode *HostFlag_Unordered  = _GetHostFlag(Res, FCMP_FLAG_UNORDERED);
-
-  SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(HostFlag_CF);
-  SetRFLAG<FEXCore::X86State::RFLAG_ZF_LOC>(HostFlag_ZF);
-  SetRFLAG<FEXCore::X86State::RFLAG_PF_LOC>(HostFlag_Unordered);
-
-  auto ZeroConst = _Constant(0);
-  SetRFLAG<FEXCore::X86State::RFLAG_AF_LOC>(ZeroConst);
-  SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(ZeroConst);
-  SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(ZeroConst);
+  GenerateFlags_FCMP(Op, Res, Src1, Src2);
 
   flagsOp = SelectionFlag::FCMP;
   flagsOpDest = Src1;
@@ -1559,8 +1584,8 @@ void OpDispatchBuilder::MOVQ2DQ(OpcodeArgs) {
 
   // This instruction is a bit special in that if the source is MMX then it zexts to 128bit
   if constexpr (ToXMM) {
-    Src = _VMov(Src, 16);
-    _StoreContext(FPRClass, 16, offsetof(FEXCore::Core::CPUState, xmm[Op->Dest.Data.GPR.GPR - FEXCore::X86State::REG_XMM_0][0]), Src);
+    Src = _VMov(16, Src);
+    _StoreContext(16, FPRClass, Src, offsetof(FEXCore::Core::CPUState, xmm[Op->Dest.Data.GPR.GPR - FEXCore::X86State::REG_XMM_0][0]));
   }
   else {
     // This is simple, just store the result
@@ -1649,11 +1674,151 @@ void OpDispatchBuilder::ADDSUBPOp(OpcodeArgs) {
   StoreResult(FPRClass, Op, ResAdd, -1);
 }
 
+void OpDispatchBuilder::PFNACCOp(OpcodeArgs) {
+  auto Size = GetSrcSize(Op);
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  OrderedNode *ResSubSrc{};
+  OrderedNode *ResSubDest{};
+  auto UpperSubDest = _VExtractElement(Size, 4, Dest, 1);
+  auto UpperSubSrc = _VExtractElement(Size, 4, Src, 1);
+
+  ResSubDest = _VFSub(4, 4, Dest, UpperSubDest);
+  ResSubSrc = _VFSub(4, 4, Src, UpperSubSrc);
+
+  auto Result = _VInsElement(8, 4, 1, 0, ResSubDest, ResSubSrc);
+
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::PFPNACCOp(OpcodeArgs) {
+  auto Size = GetSrcSize(Op);
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  OrderedNode *ResAdd{};
+  OrderedNode *ResSub{};
+  auto UpperSubDest = _VExtractElement(Size, 4, Dest, 1);
+
+  ResSub = _VFSub(4, 4, Dest, UpperSubDest);
+  ResAdd = _VFAddP(Size, 4, Src, Src);
+
+  auto Result = _VInsElement(8, 4, 1, 0, ResSub, ResAdd);
+
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::PSWAPDOp(OpcodeArgs) {
+  auto Size = GetSrcSize(Op);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  auto Result = _VRev64(Size, 4, Src);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
 template
 void OpDispatchBuilder::ADDSUBPOp<4>(OpcodeArgs);
 template
 void OpDispatchBuilder::ADDSUBPOp<8>(OpcodeArgs);
 
+void OpDispatchBuilder::PI2FWOp(OpcodeArgs) {
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  size_t Size = GetDstSize(Op);
+
+  // We now need to transpose the lower 16-bits of each element together
+  // Only needing to move the upper element down in this case
+  Src = _VInsElement(Size, 2, 1, 2, Src, Src);
+
+  // Now we need to sign extend the 16bit value to 32-bit
+  Src = _VSXTL(Size, 2, Src);
+
+  // int32_t to float
+  Src = _Vector_SToF(Size, 4, Src);
+
+  StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src, Size, -1);
+}
+
+void OpDispatchBuilder::PF2IWOp(OpcodeArgs) {
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  size_t Size = GetDstSize(Op);
+
+  // Float to int32_t
+  Src = _Vector_FToZS(Size, 4, Src);
+
+  // We now need to transpose the lower 16-bits of each element together
+  // Only needing to move the upper element down in this case
+  Src = _VInsElement(Size, 2, 1, 2, Src, Src);
+
+  // Now we need to sign extend the 16bit value to 32-bit
+  Src = _VSXTL(Size, 2, Src);
+  StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src, Size, -1);
+}
+
+void OpDispatchBuilder::PMULHRWOp(OpcodeArgs) {
+  auto Size = GetSrcSize(Op);
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+
+  OrderedNode *Res{};
+
+  // Implementation is more efficient for 8byte registers
+  // Multiplies 4 16bit values in to 4 32bit values
+  Res = _VSMull(Size * 2, 2, Dest, Src);
+  //TODO: We could remove this VCastFromGOR + VInsGPR pair if we had a VDUPFromGPR instruction that maps directly to AArch64.
+  auto M = _Constant(0x0000'8000'0000'8000ULL);
+  OrderedNode *VConstant = _VCastFromGPR(16, 8, M);
+  VConstant = _VInsGPR(16, 8, 1, VConstant, M);
+
+  Res = _VAdd(Size * 2, 4, Res, VConstant);
+
+  // Now shift and narrow to convert 32-bit values to 16bit, storing the top 16bits
+  Res = _VUShrNI(Size * 2, 4, Res, 16);
+
+  StoreResult(FPRClass, Op, Res, -1);
+}
+
+template<uint8_t CompType>
+void OpDispatchBuilder::VPFCMPOp(OpcodeArgs) {
+  auto Size = GetSrcSize(Op);
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  OrderedNode *Dest = LoadSource_WithOpSize(FPRClass, Op, Op->Dest, GetDstSize(Op), Op->Flags, -1);
+
+  OrderedNode *Result{};
+  // This maps 1:1 to an AArch64 NEON Op
+  //auto ALUOp = _VCMPGT(Size, 4, Dest, Src);
+  LogMan::Msg::DFmt("CompType: {} Size: {}", CompType, Size);
+  switch (CompType) {
+    case 0x00: // EQ
+      Result = _VFCMPEQ(Size, 4, Dest, Src);
+    break;
+    case 0x01: // GE(Swapped operand)
+      Result = _VFCMPLE(Size, 4, Src, Dest);
+    break;
+    case 0x02: // GT
+      Result = _VFCMPGT(Size, 4, Dest, Src);
+    break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Comparison type: {}", CompType);
+    break;
+  }
+
+  StoreResult(FPRClass, Op, Result, -1);
+  ShouldDump = true;
+}
+
+template
+void OpDispatchBuilder::VPFCMPOp<0>(OpcodeArgs);
+template
+void OpDispatchBuilder::VPFCMPOp<1>(OpcodeArgs);
+template
+void OpDispatchBuilder::VPFCMPOp<2>(OpcodeArgs);
+
 void OpDispatchBuilder::PMADDWD(OpcodeArgs) {
   // This is a pretty curious operation
   // Does two MADD operations across 4 16bit signed integers and accumulates to 32bit integers in the destination
@@ -1806,7 +1971,7 @@ void OpDispatchBuilder::PMULHRSW(OpcodeArgs) {
     // Implementation is more efficient for 8byte registers
     Res = _VSMull(Size * 2, 2, Dest, Src);
     Res = _VSShrI(Size * 2, 4, Res, 14);
-    auto OneVector = _VectorImm(1, Size * 2, 4);
+    auto OneVector = _VectorImm(Size * 2, 4, 1);
     Res = _VAdd(Size * 2, 4, Res, OneVector);
     Res = _VUShrNI(Size * 2, 4, Res, 1);
   }
@@ -1820,7 +1985,7 @@ void OpDispatchBuilder::PMULHRSW(OpcodeArgs) {
 
     ResultLow = _VSShrI(Size, 4, ResultLow, 14);
     ResultHigh = _VSShrI(Size, 4, ResultHigh, 14);
-    auto OneVector = _VectorImm(1, Size, 4);
+    auto OneVector = _VectorImm(Size, 4, 1);
 
     ResultLow = _VAdd(Size, 4, ResultLow, OneVector);
     ResultHigh = _VAdd(Size, 4, ResultHigh, OneVector);
@@ -2075,10 +2240,10 @@ void OpDispatchBuilder::ExtendVectorElements(OpcodeArgs) {
        CurrentElementSize != DstElementSize;
        CurrentElementSize <<= 1) {
     if constexpr (Signed) {
-      Result = _VSXTL(Result, Size, CurrentElementSize);
+      Result = _VSXTL(Size, CurrentElementSize, Result);
     }
     else {
-      Result = _VUXTL(Result, Size, CurrentElementSize);
+      Result = _VUXTL(Size, CurrentElementSize, Result);
     }
   }
   StoreResult(FPRClass, Op, Result, -1);
@@ -2132,7 +2297,7 @@ void OpDispatchBuilder::VectorRound(OpcodeArgs) {
     FEXCore::IR::Round_Host,
   };
 
-  Src = _Vector_FToI(Src, SourceModes[(RoundControlSource << 2) | RoundControl], Size, ElementSize);
+  Src = _Vector_FToI(Size, ElementSize, Src, SourceModes[(RoundControlSource << 2) | RoundControl]);
 
   if constexpr (Scalar) {
     // Insert the lower bits
@@ -2186,13 +2351,14 @@ void OpDispatchBuilder::VectorVariableBlend(OpcodeArgs) {
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
 
   // The mask is hardcoded to be xmm0 in this instruction
-  OrderedNode *Mask = _LoadContext(16, offsetof(FEXCore::Core::CPUState, xmm[0]), FPRClass);
+  OrderedNode *Mask = _LoadContext(16, FPRClass, offsetof(FEXCore::Core::CPUState, xmm[0]));
   // Each element is selected by the high bit of that element size
   // Dest[ElementIdx] = Xmm0[ElementIndex][HighBit] ? Src : Dest;
   //
   // To emulate this on AArch64
   // Arithmetic shift right by the element size, then use BSL to select the registers
   Mask = _VSShrI(Size, ElementSize, Mask, (ElementSize * 8) - 1);
+
   auto Result = _VBSL(Mask, Src, Dest);
 
   StoreResult(FPRClass, Op, Result, -1);
@@ -2205,13 +2371,16 @@ template
 void OpDispatchBuilder::VectorVariableBlend<8>(OpcodeArgs);
 
 void OpDispatchBuilder::PTestOp(OpcodeArgs) {
+  // Invalidate deferred flags early
+  InvalidateDeferredFlags();
+
   auto Size = GetSrcSize(Op);
 
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
 
-  OrderedNode *Test1 = _VAnd(Dest, Src, Size, 1);
-  OrderedNode *Test2 = _VBic(Src, Dest, Size, 1);
+  OrderedNode *Test1 = _VAnd(Size, 1, Dest, Src);
+  OrderedNode *Test2 = _VBic(Size, 1, Src, Dest);
 
   Test1 = _VPopcount(Size, 1, Test1);
   Test2 = _VPopcount(Size, 1, Test2);
@@ -2233,8 +2402,14 @@ void OpDispatchBuilder::PTestOp(OpcodeArgs) {
   Test2 = _Select(FEXCore::IR::COND_EQ,
       Test2, ZeroConst, OneConst, ZeroConst);
 
+  // Careful, these flags are different between {V,}PTEST and VTESTP{S,D}
   SetRFLAG<FEXCore::X86State::RFLAG_ZF_LOC>(Test1);
   SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(Test2);
+
+  SetRFLAG<FEXCore::X86State::RFLAG_AF_LOC>(ZeroConst);
+  SetRFLAG<FEXCore::X86State::RFLAG_SF_LOC>(ZeroConst);
+  SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(ZeroConst);
+  SetRFLAG<FEXCore::X86State::RFLAG_PF_LOC>(ZeroConst);
 }
 
 void OpDispatchBuilder::PHMINPOSUWOp(OpcodeArgs) {
@@ -2267,10 +2442,10 @@ void OpDispatchBuilder::PHMINPOSUWOp(OpcodeArgs) {
   }
 
   // Insert the minimum in to bits [15:0]
-  OrderedNode *Result = _VMov(Min, 2);
+  OrderedNode *Result = _VMov(2, Min);
 
   // Insert position in to bits [18:16]
-  Result = _VInsGPR(16, 2, Result, Pos, 1);
+  Result = _VInsGPR(16, 2, 1, Result, Pos);
 
   StoreResult(FPRClass, Op, Result, -1);
 }

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

@@ -25,12 +25,12 @@ class OrderedNode;
 OrderedNode *OpDispatchBuilder::GetX87Top() {
   // Yes, we are storing 3 bits in a single flag register.
   // Deal with it
-  return _LoadContext(1, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC, GPRClass);
+  return _LoadContext(1, GPRClass, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
 }
 
 void OpDispatchBuilder::SetX87TopTag(OrderedNode *Value, X87Tag Tag) {
   // if we are popping then we must first mark this location as empty
-  auto FTW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FTW), GPRClass);
+  auto FTW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FTW));
   OrderedNode *Mask = _Constant(0b11);
   auto TopOffset = _Lshl(Value, _Constant(1));
   Mask = _Lshl(Mask, TopOffset);
@@ -40,11 +40,11 @@ void OpDispatchBuilder::SetX87TopTag(OrderedNode *Value, X87Tag Tag) {
     NewFTW = _Or(NewFTW, TagVal);
   }
 
-  _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FTW), NewFTW);
+  _StoreContext(2, GPRClass, NewFTW, offsetof(FEXCore::Core::CPUState, FTW));
 }
 
 OrderedNode *OpDispatchBuilder::GetX87FTW(OrderedNode *Value) {
-  auto FTW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FTW), GPRClass);
+  auto FTW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FTW));
   OrderedNode *Mask = _Constant(0b11);
   auto TopOffset = _Lshl(Value, _Constant(1));
   auto NewFTW = _Lshr(FTW, TopOffset);
@@ -52,7 +52,7 @@ OrderedNode *OpDispatchBuilder::GetX87FTW(OrderedNode *Value) {
 }
 
 void OpDispatchBuilder::SetX87Top(OrderedNode *Value) {
-  _StoreContext(GPRClass, 1, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC, Value);
+  _StoreContext(1, GPRClass, Value, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
 }
 
 template<size_t width>
@@ -136,7 +136,7 @@ void OpDispatchBuilder::FLD_Const(OpcodeArgs) {
   auto low = _Constant(Lower);
   auto high = _Constant(Upper);
   OrderedNode *data = _VCastFromGPR(16, 8, low);
-  data = _VInsGPR(16, 8, data, high, 1);
+  data = _VInsGPR(16, 8, 1, data, high);
   // Write to ST[TOP]
   _StoreContextIndexed(data, top, 16, MMBaseOffset(), 16, FPRClass);
 }
@@ -202,7 +202,7 @@ void OpDispatchBuilder::FST(OpcodeArgs) {
     StoreResult_WithOpSize(FPRClass, Op, Op->Dest, data, 10, 1);
   }
   else if constexpr (width == 32 || width == 64) {
-    auto result = _F80CVT(data, width / 8);
+    auto result = _F80CVT(width / 8, data);
     StoreResult_WithOpSize(FPRClass, Op, Op->Dest, result, width / 8, 1);
   }
 
@@ -228,7 +228,7 @@ void OpDispatchBuilder::FIST(OpcodeArgs) {
 
   auto orig_top = GetX87Top();
   OrderedNode *data = _LoadContextIndexed(orig_top, 16, MMBaseOffset(), 16, FPRClass);
-  data = _F80CVTInt(data, Truncate, Size);
+  data = _F80CVTInt(Size, data, Truncate);
 
   StoreResult_WithOpSize(GPRClass, Op, Op->Dest, data, Size, 1);
 
@@ -547,9 +547,9 @@ void OpDispatchBuilder::FCHS(OpcodeArgs) {
   auto low = _Constant(0);
   auto high = _Constant(0b1'000'0000'0000'0000ULL);
   OrderedNode *data = _VCastFromGPR(16, 8, low);
-  data = _VInsGPR(16, 8, data, high, 1);
+  data = _VInsGPR(16, 8, 1, data, high);
 
-  auto result = _VXor(a, data, 16, 1);
+  auto result = _VXor(16, 1, a, data);
 
   // Write to ST[TOP]
   _StoreContextIndexed(result, top, 16, MMBaseOffset(), 16, FPRClass);
@@ -562,9 +562,9 @@ void OpDispatchBuilder::FABS(OpcodeArgs) {
   auto low = _Constant(~0ULL);
   auto high = _Constant(0b0'111'1111'1111'1111ULL);
   OrderedNode *data = _VCastFromGPR(16, 8, low);
-  data = _VInsGPR(16, 8, data, high, 1);
+  data = _VInsGPR(16, 8, 1, data, high);
 
-  auto result = _VAnd(a, data, 16, 1);
+  auto result = _VAnd(16, 1, a, data);
 
   // Write to ST[TOP]
   _StoreContextIndexed(result, top, 16, MMBaseOffset(), 16, FPRClass);
@@ -621,10 +621,10 @@ 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(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FCW), NewFCW);
+  _StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 
   // Init FSW to 0
   SetX87Top(_Constant(0));
@@ -635,7 +635,7 @@ void OpDispatchBuilder::FNINIT(OpcodeArgs) {
   SetRFLAG<FEXCore::X86State::X87FLAG_C3_LOC>(_Constant(0));
 
   // Tags all get set to 0b11
-  _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FTW), _Constant(0xFFFF));
+  _StoreContext(2, GPRClass, _Constant(0xFFFF), offsetof(FEXCore::Core::CPUState, FTW));
 }
 
 template<size_t width, bool Integer, OpDispatchBuilder::FCOMIFlags whichflags, bool poptwice>
@@ -685,12 +685,15 @@ void OpDispatchBuilder::FCOMI(OpcodeArgs) {
     SetRFLAG<FEXCore::X86State::X87FLAG_C3_LOC>(HostFlag_ZF);
   }
   else {
+    // Invalidate deferred flags early
+    // OF, SF, AF, PF all undefined
+    InvalidateDeferredFlags();
+
     SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(HostFlag_CF);
     SetRFLAG<FEXCore::X86State::RFLAG_ZF_LOC>(HostFlag_ZF);
     SetRFLAG<FEXCore::X86State::RFLAG_PF_LOC>(HostFlag_Unordered);
   }
 
-
   if constexpr (poptwice) {
     // if we are popping then we must first mark this location as empty
     SetX87TopTag(top, X87Tag::Empty);
@@ -872,7 +875,7 @@ void OpDispatchBuilder::X87FYL2X(OpcodeArgs) {
     auto low = _Constant(0x8000'0000'0000'0000ULL);
     auto high = _Constant(0b0'011'1111'1111'1111);
     OrderedNode *data = _VCastFromGPR(16, 8, low);
-    data = _VInsGPR(16, 8, data, high, 1);
+    data = _VInsGPR(16, 8, 1, data, high);
     st0 = _F80Add(st0, data);
   }
 
@@ -895,7 +898,7 @@ void OpDispatchBuilder::X87TAN(OpcodeArgs) {
   auto low = _Constant(0x8000'0000'0000'0000ULL);
   auto high = _Constant(0b0'011'1111'1111'1111ULL);
   OrderedNode *data = _VCastFromGPR(16, 8, low);
-  data = _VInsGPR(16, 8, data, high, 1);
+  data = _VInsGPR(16, 8, 1, data, high);
 
   // Write to ST[TOP]
   _StoreContextIndexed(result, orig_top, 16, MMBaseOffset(), 16, FPRClass);
@@ -925,7 +928,7 @@ void OpDispatchBuilder::X87LDENV(OpcodeArgs) {
 
   auto NewFCW = _LoadMem(GPRClass, 2, Mem, 2);
   _F80LoadFCW(NewFCW);
-  _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FCW), NewFCW);
+  _StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 
   OrderedNode *MemLocation = _Add(Mem, _Constant(Size * 1));
   auto NewFSW = _LoadMem(GPRClass, Size, MemLocation, Size);
@@ -948,7 +951,7 @@ void OpDispatchBuilder::X87LDENV(OpcodeArgs) {
     // FTW
     OrderedNode *MemLocation = _Add(Mem, _Constant(Size * 2));
     auto NewFTW = _LoadMem(GPRClass, Size, MemLocation, Size);
-    _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FTW), NewFTW);
+    _StoreContext(2, GPRClass, NewFTW, offsetof(FEXCore::Core::CPUState, FTW));
   }
 }
 
@@ -977,7 +980,7 @@ void OpDispatchBuilder::X87FNSTENV(OpcodeArgs) {
   Mem = AppendSegmentOffset(Mem, Op->Flags);
 
   {
-    auto FCW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FCW), GPRClass);
+    auto FCW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
     _StoreMem(GPRClass, Size, Mem, FCW, Size);
   }
 
@@ -1005,7 +1008,7 @@ void OpDispatchBuilder::X87FNSTENV(OpcodeArgs) {
   {
     // FTW
     OrderedNode *MemLocation = _Add(Mem, _Constant(Size * 2));
-    auto FTW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FTW), GPRClass);
+    auto FTW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FTW));
     _StoreMem(GPRClass, Size, MemLocation, FTW, Size);
   }
 
@@ -1037,11 +1040,11 @@ void OpDispatchBuilder::X87FNSTENV(OpcodeArgs) {
 void OpDispatchBuilder::X87FLDCW(OpcodeArgs) {
   OrderedNode *NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, -1);
   _F80LoadFCW(NewFCW);
-  _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FCW), NewFCW);
+  _StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 }
 
 void OpDispatchBuilder::X87FSTCW(OpcodeArgs) {
-  auto FCW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FCW), GPRClass);
+  auto FCW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
 
   StoreResult(GPRClass, Op, FCW, -1);
 }
@@ -1108,7 +1111,7 @@ void OpDispatchBuilder::X87FNSAVE(OpcodeArgs) {
 
   OrderedNode *Top = GetX87Top();
   {
-    auto FCW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FCW), GPRClass);
+    auto FCW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
     _StoreMem(GPRClass, Size, Mem, FCW, Size);
   }
 
@@ -1135,7 +1138,7 @@ void OpDispatchBuilder::X87FNSAVE(OpcodeArgs) {
   {
     // FTW
     OrderedNode *MemLocation = _Add(Mem, _Constant(Size * 2));
-    auto FTW = _LoadContext(2, offsetof(FEXCore::Core::CPUState, FTW), GPRClass);
+    auto FTW = _LoadContext(2, GPRClass, offsetof(FEXCore::Core::CPUState, FTW));
     _StoreMem(GPRClass, Size, MemLocation, FTW, Size);
   }
 
@@ -1196,7 +1199,7 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
 
   auto NewFCW = _LoadMem(GPRClass, 2, Mem, 2);
   _F80LoadFCW(NewFCW);
-  _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FCW), NewFCW);
+  _StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 
   OrderedNode *MemLocation = _Add(Mem, _Constant(Size * 1));
   auto NewFSW = _LoadMem(GPRClass, Size, MemLocation, Size);
@@ -1219,7 +1222,7 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
     // FTW
     OrderedNode *MemLocation = _Add(Mem, _Constant(Size * 2));
     auto NewFTW = _LoadMem(GPRClass, Size, MemLocation, Size);
-    _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FTW), NewFTW);
+    _StoreContext(2, GPRClass, NewFTW, offsetof(FEXCore::Core::CPUState, FTW));
   }
 
   OrderedNode *ST0Location = _Add(Mem, _Constant(Size * 7));
@@ -1231,7 +1234,7 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
   auto low = _Constant(~0ULL);
   auto high = _Constant(0xFFFF);
   OrderedNode *Mask = _VCastFromGPR(16, 8, low);
-  Mask = _VInsGPR(16, 8, Mask, high, 1);
+  Mask = _VInsGPR(16, 8, 1, Mask, high);
 
   for (int i = 0; i < 7; ++i) {
     OrderedNode *Reg = _LoadMem(FPRClass, 16, ST0Location, 1);
@@ -1359,7 +1362,7 @@ void OpDispatchBuilder::X87FCMOV(OpcodeArgs) {
   SrcCond = _Sbfe(1, 0, SrcCond);
 
   OrderedNode *VecCond = _VCastFromGPR(16, 8, SrcCond);
-  VecCond = _VInsGPR(16, 8, VecCond, SrcCond, 1);
+  VecCond = _VInsGPR(16, 8, 1, VecCond, SrcCond);
 
   auto top = GetX87Top();
   OrderedNode* arg;
@@ -1380,7 +1383,7 @@ void OpDispatchBuilder::X87FCMOV(OpcodeArgs) {
 
 void OpDispatchBuilder::X87EMMS(OpcodeArgs) {
   // Tags all get set to 0b11
-  _StoreContext(GPRClass, 2, offsetof(FEXCore::Core::CPUState, FTW), _Constant(0xFFFF));
+  _StoreContext(2, GPRClass, _Constant(0xFFFF), offsetof(FEXCore::Core::CPUState, FTW));
 }
 
 void OpDispatchBuilder::X87FFREE(OpcodeArgs) {

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/DDDTables.cpp

@@ -15,37 +15,42 @@ using namespace InstFlags;
 
 void InitializeDDDTables() {
   static constexpr U8U8InfoStruct DDDNowOpTable[] = {
-    {0x0C, 1, X86InstInfo{"PI2FW",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0x0D, 1, X86InstInfo{"PI2FD",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0x1C, 1, X86InstInfo{"PF2IW",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0x1D, 1, X86InstInfo{"PF2ID",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
+    {0x0C, 1, X86InstInfo{"PI2FW",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x0D, 1, X86InstInfo{"PI2FD",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x1C, 1, X86InstInfo{"PF2IW",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x1D, 1, X86InstInfo{"PF2ID",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
 
-    {0x8A, 1, X86InstInfo{"PFNACC",   TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0x8E, 1, X86InstInfo{"PFPNACC",  TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
+    // Inverse 3DNow! These two instructions are Geode product line specific
+    // No CPUID for these, you're expected to read ID_CONFIG_MSR (1250h) bit 1
+    {0x86, 1, X86InstInfo{"PFRCPV",   TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x87, 1, X86InstInfo{"PFRSQRTV", TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
 
-    {0x9A, 1, X86InstInfo{"PFSUB",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0x9E, 1, X86InstInfo{"PFADD",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
+    {0x8A, 1, X86InstInfo{"PFNACC",   TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x8E, 1, X86InstInfo{"PFPNACC",  TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
 
-    {0xAA, 1, X86InstInfo{"PFSUBR",   TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0xAE, 1, X86InstInfo{"PFACC",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
+    {0x90, 1, X86InstInfo{"PFCMPGE",  TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x94, 1, X86InstInfo{"PFMIN",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x96, 1, X86InstInfo{"PFRCP",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x97, 1, X86InstInfo{"PFRSQRT",  TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
 
-    {0xBB, 1, X86InstInfo{"PSWAPD",   TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0xBF, 1, X86InstInfo{"PAVGUSB",  TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
+    {0x9A, 1, X86InstInfo{"PFSUB",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0x9E, 1, X86InstInfo{"PFADD",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
 
-    {0x90, 1, X86InstInfo{"PFCMPGE",  TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0x94, 1, X86InstInfo{"PFMIN",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0x96, 1, X86InstInfo{"PFRCP",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0x97, 1, X86InstInfo{"PFRSQRT",  TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
+    {0xA0, 1, X86InstInfo{"PFCMPGT",  TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0xA4, 1, X86InstInfo{"PFMAX",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0xA6, 1, X86InstInfo{"PFRCPIT1", TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0xA7, 1, X86InstInfo{"PFRSQIT1", TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
 
-    {0xA0, 1, X86InstInfo{"PFCMPGT",  TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0xA4, 1, X86InstInfo{"PFMAX",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0xA6, 1, X86InstInfo{"PFRCPIT1", TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0xA7, 1, X86InstInfo{"PFRSQIT1", TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
+    {0xAA, 1, X86InstInfo{"PFSUBR",   TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0xAE, 1, X86InstInfo{"PFACC",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
 
-    {0xB0, 1, X86InstInfo{"PFCMPEQ",  TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0xB4, 1, X86InstInfo{"PFMUL",    TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0xB6, 1, X86InstInfo{"PFRCPIT2", TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
-    {0xB7, 1, X86InstInfo{"PMULHRW",  TYPE_3DNOW_INST, FLAGS_MODRM, 0, nullptr}},
+    {0xB0, 1, X86InstInfo{"PFCMPEQ",  TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0xB4, 1, X86InstInfo{"PFMUL",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0xB6, 1, X86InstInfo{"PFRCPIT2", TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0xB7, 1, X86InstInfo{"PMULHRW",  TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+
+    {0xBB, 1, X86InstInfo{"PSWAPD",   TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
+    {0xBF, 1, X86InstInfo{"PAVGUSB",  TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
   };
 
   GenerateTable(&DDDNowOps.at(0), DDDNowOpTable, std::size(DDDNowOpTable));

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

@@ -14,11 +14,11 @@ namespace FEXCore::X86Tables {
 using namespace InstFlags;
 
 void InitializeH0F38Tables() {
-#define OPD(prefix, opcode) ((prefix << 8) | opcode)
+#define OPD(prefix, opcode) (((prefix) << 8) | opcode)
   constexpr uint16_t PF_38_NONE = 0;
-  constexpr uint16_t PF_38_66   = 1;
-  constexpr uint16_t PF_38_F2   = 2;
-  constexpr uint16_t PF_38_F3   = 3;
+  constexpr uint16_t PF_38_66   = (1U << 0);
+  constexpr uint16_t PF_38_F2   = (1U << 1);
+  constexpr uint16_t PF_38_F3   = (1U << 2);
 
   static constexpr U16U8InfoStruct H0F38Table[] = {
     {OPD(PF_38_NONE, 0x00), 1, X86InstInfo{"PSHUFB",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT)  | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 0, nullptr}},
@@ -74,6 +74,7 @@ void InitializeH0F38Tables() {
     {OPD(PF_38_66,   0x33), 1, X86InstInfo{"PMOVZXWD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(PF_38_66,   0x34), 1, X86InstInfo{"PMOVZXWQ",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(PF_38_66,   0x35), 1, X86InstInfo{"PMOVZXDQ",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(PF_38_66,   0x37), 1, X86InstInfo{"PCMPGTQ",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(PF_38_66,   0x38), 1, X86InstInfo{"PMINSB",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(PF_38_66,   0x39), 1, X86InstInfo{"PMINSD",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(PF_38_66,   0x3A), 1, X86InstInfo{"PMINUW",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
@@ -98,8 +99,10 @@ void InitializeH0F38Tables() {
     {OPD(PF_38_66, 0xF0), 1, X86InstInfo{"MOVBE",      TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
     {OPD(PF_38_66, 0xF1), 1, X86InstInfo{"MOVBE",      TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
 
-    {OPD(PF_38_F2,   0xF0), 1, X86InstInfo{"CRC32",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(PF_38_F2,   0xF1), 1, X86InstInfo{"CRC32",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(PF_38_F2,   0xF0), 1, X86InstInfo{"CRC32",      TYPE_INST, GenFlagsSizes(SIZE_DEF, SIZE_8BIT) | FLAGS_MODRM, 0, nullptr}},
+    {OPD(PF_38_F2,   0xF1), 1, X86InstInfo{"CRC32",      TYPE_INST, FLAGS_MODRM, 0, nullptr}},
+    {OPD(PF_38_66 | PF_38_F2,   0xF0), 1, X86InstInfo{"CRC32",      TYPE_INST, GenFlagsSizes(SIZE_DEF, SIZE_8BIT) | FLAGS_MODRM, 0, nullptr}},
+    {OPD(PF_38_66 | PF_38_F2,   0xF1), 1, X86InstInfo{"CRC32",      TYPE_INST, FLAGS_MODRM, 0, nullptr}},
 
     {OPD(PF_38_66,   0xF6), 1, X86InstInfo{"ADCX",       TYPE_INST, FLAGS_MODRM, 0, nullptr}},
     {OPD(PF_38_F3,   0xF6), 1, X86InstInfo{"ADOX",       TYPE_INST, FLAGS_MODRM, 0, nullptr}},

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

@@ -144,38 +144,38 @@ void InitializeSecondaryGroupTables() {
 
     // AMD documentation is a bit broken for Group 9
     // Claims the entire group has n/a applied for the prefix (Implies that the prefix is ignored)
-    // RDRAND/RDSEED only work with no prefix
+    // RDRAND/RDSEED only work with no prefix (Other than 66h)
     // CMPXCHG8B/16B works with all prefixes
     // Tooling fails to decode CMPXCHG with prefix
     {OPD(TYPE_GROUP_9, PF_NONE, 0), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,   0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_NONE, 1), 1, X86InstInfo{"CMPXCHG16B", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_NONE, 1), 1, X86InstInfo{"CMPXCHG8B/16B", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_NONE, 2), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_NONE, 3), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_NONE, 4), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_NONE, 5), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,   0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_NONE, 6), 1, X86InstInfo{"RDRAND",     TYPE_UNDEC, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_REG_ONLY, 0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_NONE, 7), 1, X86InstInfo{"RDSEED",     TYPE_UNDEC, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_REG_ONLY, 0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_NONE, 6), 1, X86InstInfo{"RDRAND",     TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_REG_ONLY, 0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_NONE, 7), 1, X86InstInfo{"RDSEED",     TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_REG_ONLY, 0, nullptr}},
 
     {OPD(TYPE_GROUP_9, PF_F3, 0), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_F3, 1), 1, X86InstInfo{"CMPXCHG16B", TYPE_INVALID, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY,   0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_F3, 1), 1, X86InstInfo{"CMPXCHG8B/16B", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_F3, 2), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_F3, 3), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_F3, 4), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_F3, 5), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_F3, 6), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_F3, 7), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_F3, 7), 1, X86InstInfo{"RDPID",      TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
 
     {OPD(TYPE_GROUP_9, PF_66, 0), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_66, 1), 1, X86InstInfo{"CMPXCHG16B", TYPE_INVALID, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY,   0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_66, 1), 1, X86InstInfo{"CMPXCHG8B/16B", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_66, 2), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_66, 3), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_66, 4), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_66, 5), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_66, 6), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_66, 7), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_66, 6), 1, X86InstInfo{"RDRAND",     TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_REG_ONLY, 0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_66, 7), 1, X86InstInfo{"RDSEED",     TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_REG_ONLY, 0, nullptr}},
 
     {OPD(TYPE_GROUP_9, PF_F2, 0), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
-    {OPD(TYPE_GROUP_9, PF_F2, 1), 1, X86InstInfo{"CMPXCHG16B", TYPE_INVALID, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY,   0, nullptr}},
+    {OPD(TYPE_GROUP_9, PF_F2, 1), 1, X86InstInfo{"CMPXCHG8B/16B", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_F2, 2), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_F2, 3), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
     {OPD(TYPE_GROUP_9, PF_F2, 4), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},

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

@@ -120,9 +120,9 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0x6F, 1, X86InstInfo{"MOVQ",       TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX,                                   0, nullptr}},
 
     {0x70, 1, X86InstInfo{"PSHUFW",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX,                                   1, nullptr}},
-    {0x71, 1, X86InstInfo{"",           TYPE_GROUP_12, FLAGS_NONE,                                                                                      0, nullptr}},
-    {0x72, 1, X86InstInfo{"",           TYPE_GROUP_13, FLAGS_NONE,                                                                                      0, nullptr}},
-    {0x73, 1, X86InstInfo{"",           TYPE_GROUP_14, FLAGS_NONE,                                                                                      0, nullptr}},
+    {0x71, 1, X86InstInfo{"",           TYPE_GROUP_12, FLAGS_NO_OVERLAY,                                                                                0, nullptr}},
+    {0x72, 1, X86InstInfo{"",           TYPE_GROUP_13, FLAGS_NO_OVERLAY,                                                                                0, nullptr}},
+    {0x73, 1, X86InstInfo{"",           TYPE_GROUP_14, FLAGS_NO_OVERLAY,                                                                                0, nullptr}},
     {0x74, 1, X86InstInfo{"PCMPEQB",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX,                         0, nullptr}},
     {0x75, 1, X86InstInfo{"PCMPEQW",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX,                         0, nullptr}},
     {0x76, 1, X86InstInfo{"PCMPEQD",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX,                         0, nullptr}},
@@ -186,8 +186,8 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0xB6, 1, X86InstInfo{"MOVZX",   TYPE_INST, GenFlagsSrcSize(SIZE_8BIT) | FLAGS_MODRM | FLAGS_NO_OVERLAY,                                        0, nullptr}},
     {0xB7, 1, X86InstInfo{"MOVZX",   TYPE_INST, GenFlagsSrcSize(SIZE_16BIT) | FLAGS_MODRM | FLAGS_NO_OVERLAY,                                       0, nullptr}},
     {0xB8, 1, X86InstInfo{"",        TYPE_INVALID, FLAGS_NONE,                                                                                      0, nullptr}},
-    {0xB9, 1, X86InstInfo{"",        TYPE_GROUP_10, FLAGS_NONE,                                                                                     0, nullptr}},
-    {0xBA, 1, X86InstInfo{"",        TYPE_GROUP_8, FLAGS_NONE,                                                                                      0, nullptr}},
+    {0xB9, 1, X86InstInfo{"",        TYPE_GROUP_10, FLAGS_NO_OVERLAY,                                                                               0, nullptr}},
+    {0xBA, 1, X86InstInfo{"",        TYPE_GROUP_8, FLAGS_NO_OVERLAY,                                                                                0, nullptr}},
     {0xBB, 1, X86InstInfo{"BTC",     TYPE_INST, FLAGS_DEBUG_MEM_ACCESS | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_NO_OVERLAY,                         0, nullptr}},
     {0xBC, 1, X86InstInfo{"BSF",     TYPE_INST, FLAGS_MODRM | FLAGS_NO_OVERLAY66,                                                                   0, nullptr}},
     {0xBD, 1, X86InstInfo{"BSR",     TYPE_INST, FLAGS_MODRM | FLAGS_NO_OVERLAY66,                                                                   0, nullptr}},
@@ -201,7 +201,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0xC4, 1, X86InstInfo{"PINSRW",  TYPE_INST, GenFlagsSizes(SIZE_64BIT, SIZE_16BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX | FLAGS_SF_SRC_GPR,           1, nullptr}},
     {0xC5, 1, X86InstInfo{"PEXTRW",  TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 1, nullptr}},
     {0xC6, 1, X86InstInfo{"SHUFPS",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                                                     1, nullptr}},
-    {0xC7, 1, X86InstInfo{"",        TYPE_GROUP_9, FLAGS_NONE,                                                                                                     0, nullptr}},
+    {0xC7, 1, X86InstInfo{"",        TYPE_GROUP_9, FLAGS_NO_OVERLAY,                                                                                               0, nullptr}},
     {0xC8, 8, X86InstInfo{"BSWAP",   TYPE_INST, FLAGS_SF_REX_IN_BYTE | FLAGS_NO_OVERLAY,                                                                           0, nullptr}},
 
     {0xD0, 1, X86InstInfo{"",         TYPE_INVALID, FLAGS_NONE,                                                                                         0, nullptr}},

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

@@ -349,9 +349,17 @@ 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 (Thread->CPUBackend->NeedsRetainedIRCopy()) {
+          // 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)});
+        }
+        else {
+          // If the IR doesn't need to be retained then we can just delete it now
+          delete DebugData;
+          delete RAData;
+          delete IRList;
+        }
       }
     }
 

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

@@ -171,6 +171,19 @@ static void PrintArg(std::stringstream *out, [[maybe_unused]] IRListView const*
   }
 }
 
+static void PrintArg(std::stringstream *out, [[maybe_unused]] IRListView const* IR, FEXCore::IR::SyscallFlags Arg) {
+  switch (Arg) {
+    case FEXCore::IR::SyscallFlags::DEFAULT: *out << "Default"; break;
+    case FEXCore::IR::SyscallFlags::OPTIMIZETHROUGH: *out << "Optimize Through"; break;
+    case FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY: *out << "No Sync State on Entry"; break;
+    case FEXCore::IR::SyscallFlags::NORETURN: *out << "No Return"; break;
+    case FEXCore::IR::SyscallFlags::NOSIDEEFFECTS: *out << "No Side Effects"; break;
+    default: *out << "<Unknown Round Type>"; break;
+  }
+}
+
+
+
 void Dump(std::stringstream *out, IRListView const* IR, IR::RegisterAllocationData *RAData) {
   auto HeaderOp = IR->GetHeader();
 
@@ -225,7 +238,7 @@ void Dump(std::stringstream *out, IRListView const* IR, IR::RegisterAllocationDa
             NumElements /= ElementSize;
           }
 
-          *out << "%ssa" << ID;
+          *out << "%ssa" << std::dec << ID;
 
           if (RAData) {
             auto PhyReg = RAData->GetNodeRegister(ID);
@@ -265,7 +278,7 @@ void Dump(std::stringstream *out, IRListView const* IR, IR::RegisterAllocationDa
             NumElements = IROp->Size / ElementSize;
           }
 
-          *out << "(%ssa" << ID << ' ';
+          *out << "(%ssa" << std::dec << ID << ' ';
           *out << 'i' << std::dec << (ElementSize * 8);
           if (NumElements > 1) {
             *out << 'v' << std::dec << NumElements;

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

@@ -16,6 +16,62 @@ $end_info$
 #include <vector>
 
 namespace FEXCore::IR {
+FEXCore::IR::RegisterClassType IREmitter::WalkFindRegClass(OrderedNode *Node) {
+  auto Class = GetOpRegClass(Node);
+  switch (Class) {
+    case GPRClass:
+    case GPRPairClass:
+    case FPRClass:
+    case GPRFixedClass:
+    case FPRFixedClass:
+    case InvalidClass:
+      return Class;
+    default: break;
+  }
+
+  // Complex case, needs to be handled on an op by op basis
+  uintptr_t DataBegin = DualListData.DataBegin();
+
+  FEXCore::IR::IROp_Header *IROp = Node->Op(DataBegin);
+
+  switch (IROp->Op) {
+    case IROps::OP_LOADREGISTER: {
+      auto Op = IROp->C<IROp_LoadRegister>();
+      return Op->Class;
+      break;
+    }
+    case IROps::OP_LOADCONTEXT: {
+      auto Op = IROp->C<IROp_LoadContext>();
+      return Op->Class;
+      break;
+    }
+    case IROps::OP_LOADCONTEXTINDEXED: {
+      auto Op = IROp->C<IROp_LoadContextIndexed>();
+      return Op->Class;
+      break;
+    }
+    case IROps::OP_FILLREGISTER: {
+      auto Op = IROp->C<IROp_FillRegister>();
+      return Op->Class;
+      break;
+    }
+    case IROps::OP_LOADMEM: {
+      auto Op = IROp->C<IROp_LoadMem>();
+      return Op->Class;
+      break;
+    }
+    case IROps::OP_LOADMEMTSO: {
+      auto Op = IROp->C<IROp_LoadMemTSO>();
+      return Op->Class;
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled op type: {} {} in argument class validation", IROp->Op, GetOpName(Node));
+      break;
+  }
+  return InvalidClass;
+}
+
 void IREmitter::ResetWorkingList() {
   DualListData.Reset();
   CodeBlocks.clear();

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";
@@ -170,7 +150,7 @@ class IRParser: public FEXCore::IR::IREmitter {
       }
       Result.data[i] = high * 16 + low;
     }
-    
+
     return {DecodeFailure::DECODE_OKAY, Result};
   }
 
@@ -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) {
@@ -351,10 +331,6 @@ class IRParser: public FEXCore::IR::IREmitter {
 
   bool Loaded = false;
 
-#define IROP_PARSER_ALLOCATE_HELPERS
-#include <FEXCore/IR/IRDefines.inc>
-
-
   bool Parse() {
     const auto CheckPrintError = [&](const LineDefinition &Def, DecodeFailure Failure) -> bool {
       if (Failure != DecodeFailure::DECODE_OKAY) {
@@ -367,6 +343,18 @@ class IRParser: public FEXCore::IR::IREmitter {
       return true;
     };
 
+    const auto CheckPrintErrorArg = [&](const LineDefinition &Def, DecodeFailure Failure, size_t Arg) -> bool {
+      if (Failure != DecodeFailure::DECODE_OKAY) {
+        LogMan::Msg::EFmt("Error on Line: {}", Def.LineNumber);
+        LogMan::Msg::EFmt("{}", Lines[Def.LineNumber]);
+        LogMan::Msg::EFmt("Argument Number {}: {}", Arg + 1, Def.Args[Arg]);
+        LogMan::Msg::EFmt("Value Couldn't be decoded due to {}", DecodeErrorToString(Failure));
+        return false;
+      }
+
+      return true;
+    };
+
     // String parse every line for our definitions
     for (size_t i = 0; i < Lines.size(); ++i) {
       std::string Line = Lines[i];
@@ -374,7 +362,7 @@ 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()) {
@@ -393,7 +381,7 @@ 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
         }
@@ -413,7 +401,7 @@ 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)) {
@@ -422,7 +410,7 @@ 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;
         }
@@ -439,8 +427,8 @@ 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;
@@ -457,11 +445,11 @@ 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;
       }
@@ -478,7 +466,7 @@ 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?
@@ -487,7 +475,7 @@ 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));
 

External/FEXCore/Source/Interface/IR/Passes/ConstProp.cpp

@@ -286,7 +286,7 @@ void ConstProp::FCMPOptimization(IREmitter *IREmit, const IRListView& CurrentIR)
     if (IROp->Op == OP_GETHOSTFLAG) {
       auto ghf = IROp->CW<IR::IROp_GetHostFlag>();
 
-      auto fcmp = IREmit->GetOpHeader(ghf->GPR)->CW<IR::IROp_FCmp>();
+      auto fcmp = IREmit->GetOpHeader(ghf->Value)->CW<IR::IROp_FCmp>();
       LOGMAN_THROW_A_FMT(fcmp->Header.Op == OP_FCMP || fcmp->Header.Op == OP_F80CMP, "Unexpected OP_GETHOSTFLAG source");
       if(fcmp->Header.Op == OP_FCMP) {
         fcmp->Flags |= 1 << ghf->Flag;
@@ -301,18 +301,29 @@ void ConstProp::LoadMemStoreMemImmediatePooling(IREmitter *IREmit, const IRListV
   for (auto [BlockNode, BlockIROp] : CurrentIR.GetBlocks()) {
     for (auto [CodeNode, IROp] : CurrentIR.GetCode(BlockNode)) {
       if (IROp->Op == OP_LOADMEM || IROp->Op == OP_STOREMEM) {
+        size_t AddrIndex = 0;
+        size_t OffsetIndex = 0;
+
+        if (IROp->Op == OP_LOADMEM) {
+          AddrIndex = IR::IROp_LoadMem::Addr_Index;
+          OffsetIndex = IR::IROp_LoadMem::Offset_Index;
+        }
+        else {
+          AddrIndex = IR::IROp_StoreMem::Addr_Index;
+          OffsetIndex = IR::IROp_StoreMem::Offset_Index;
+        }
         uint64_t Addr;
 
-        if (IREmit->IsValueConstant(IROp->Args[0], &Addr) && IROp->Args[1].IsInvalid()) {
+        if (IREmit->IsValueConstant(IROp->Args[AddrIndex], &Addr) && IROp->Args[OffsetIndex].IsInvalid()) {
           for (auto& Const: AddressgenConsts) {
             if ((Addr - Const.second) < 65536) {
-              IREmit->ReplaceNodeArgument(CodeNode, 0, Const.first);
-              IREmit->ReplaceNodeArgument(CodeNode, 1, IREmit->_Constant(Addr - Const.second));
+              IREmit->ReplaceNodeArgument(CodeNode, AddrIndex, Const.first);
+              IREmit->ReplaceNodeArgument(CodeNode, OffsetIndex, IREmit->_Constant(Addr - Const.second));
               goto doneOp;
             }
           }
 
-          AddressgenConsts[IREmit->UnwrapNode(IROp->Args[0])] = Addr;
+          AddressgenConsts[IREmit->UnwrapNode(IROp->Args[AddrIndex])] = Addr;
         }
         doneOp:
         ;
@@ -374,8 +385,8 @@ bool ConstProp::ZextAndMaskingElimination(IREmitter *IREmit, const IRListView& C
       auto Op = IROp->C<IR::IROp_Bfe>();
 
       // Is this value already BFE'd?
-      if (IsBfeAlreadyDone(IREmit, IROp->Args[0], Op->Width)) {
-        IREmit->ReplaceAllUsesWith(CodeNode, CurrentIR.GetNode(IROp->Args[0]));
+      if (IsBfeAlreadyDone(IREmit, Op->Src, Op->Width)) {
+        IREmit->ReplaceAllUsesWith(CodeNode, CurrentIR.GetNode(Op->Src));
         //printf("Removed BFE once \n");
         break;
       }
@@ -383,7 +394,7 @@ bool ConstProp::ZextAndMaskingElimination(IREmitter *IREmit, const IRListView& C
       // Is this value already ZEXT'd?
       if (Op->lsb == 0) {
         //LoadMem, LoadMemTSO & LoadContext ZExt
-        auto source = IROp->Args[0];
+        auto source = Op->Src;
         auto sourceHeader = IREmit->GetOpHeader(source);
 
         if (Op->Width >= (sourceHeader->Size*8)  &&
@@ -401,10 +412,10 @@ bool ConstProp::ZextAndMaskingElimination(IREmitter *IREmit, const IRListView& C
       imm = (imm-1) *2 + 1;
       imm <<= Op->lsb;
 
-      auto newArg = RemoveUselessMasking(IREmit, IROp->Args[0], imm);
+      auto newArg = RemoveUselessMasking(IREmit, Op->Src, imm);
 
-      if (newArg.ID() != IROp->Args[0].ID()) {
-        IREmit->ReplaceNodeArgument(CodeNode, 0, IREmit->UnwrapNode(newArg));
+      if (newArg.ID() != Op->Src.ID()) {
+        IREmit->ReplaceNodeArgument(CodeNode, Op->Src_Index, IREmit->UnwrapNode(newArg));
         Changed = true;
       }
       break;
@@ -418,10 +429,10 @@ bool ConstProp::ZextAndMaskingElimination(IREmitter *IREmit, const IRListView& C
       imm = (imm-1) *2 + 1;
       imm <<= Op->lsb;
 
-      auto newArg = RemoveUselessMasking(IREmit, IROp->Args[0], imm);
+      auto newArg = RemoveUselessMasking(IREmit, Op->Src, imm);
 
-      if (newArg.ID() != IROp->Args[0].ID()) {
-        IREmit->ReplaceNodeArgument(CodeNode, 0, IREmit->UnwrapNode(newArg));
+      if (newArg.ID() != Op->Src.ID()) {
+        IREmit->ReplaceNodeArgument(CodeNode, Op->Src_Index, IREmit->UnwrapNode(newArg));
         Changed = true;
       }
       break;
@@ -528,15 +539,15 @@ bool ConstProp::ConstantPropagation(IREmitter *IREmit, const IRListView& Current
 
     case OP_LOADMEM: {
       auto Op = IROp->CW<IR::IROp_LoadMem>();
-      auto AddressHeader = IREmit->GetOpHeader(Op->Header.Args[0]);
+      auto AddressHeader = IREmit->GetOpHeader(Op->Addr);
 
       if (AddressHeader->Op == OP_ADD && AddressHeader->Size == 8) {
         auto [OffsetType, OffsetScale, Arg0, Arg1] = MemExtendedAddressing(IREmit, IROp->Size, AddressHeader);
 
         Op->OffsetType = OffsetType;
         Op->OffsetScale = OffsetScale;
-        IREmit->ReplaceNodeArgument(CodeNode, 0, Arg0);
-        IREmit->ReplaceNodeArgument(CodeNode, 1, Arg1);
+        IREmit->ReplaceNodeArgument(CodeNode, Op->Addr_Index, Arg0); // Addr
+        IREmit->ReplaceNodeArgument(CodeNode, Op->Offset_Index, Arg1); // Offset
 
         Changed = true;
       }
@@ -545,15 +556,15 @@ bool ConstProp::ConstantPropagation(IREmitter *IREmit, const IRListView& Current
 
     case OP_STOREMEM: {
       auto Op = IROp->CW<IR::IROp_StoreMem>();
-      auto AddressHeader = IREmit->GetOpHeader(Op->Header.Args[0]);
+      auto AddressHeader = IREmit->GetOpHeader(Op->Addr);
 
       if (AddressHeader->Op == OP_ADD && AddressHeader->Size == 8) {
         auto [OffsetType, OffsetScale, Arg0, Arg1] = MemExtendedAddressing(IREmit, IROp->Size, AddressHeader);
 
         Op->OffsetType = OffsetType;
         Op->OffsetScale = OffsetScale;
-        IREmit->ReplaceNodeArgument(CodeNode, 0, Arg0);
-        IREmit->ReplaceNodeArgument(CodeNode, 2, Arg1);
+        IREmit->ReplaceNodeArgument(CodeNode, Op->Addr_Index, Arg0); // Addr
+        IREmit->ReplaceNodeArgument(CodeNode, Op->Offset_Index, Arg1); // Offset
 
         Changed = true;
       }
@@ -707,7 +718,7 @@ bool ConstProp::ConstantPropagation(IREmitter *IREmit, const IRListView& Current
     case OP_BFE: {
       auto Op = IROp->C<IR::IROp_Bfe>();
       uint64_t Constant;
-      if (IROp->Size <= 8 && IREmit->IsValueConstant(Op->Header.Args[0], &Constant)) {
+      if (IROp->Size <= 8 && IREmit->IsValueConstant(Op->Src, &Constant)) {
         uint64_t SourceMask = (1ULL << Op->Width) - 1;
         if (Op->Width == 64)
           SourceMask = ~0ULL;
@@ -897,7 +908,7 @@ bool ConstProp::ConstantInlining(IREmitter *IREmit, const IRListView& CurrentIR)
 
         uint64_t Constant{};
         if (IREmit->IsValueConstant(Op->NewRIP, &Constant)) {
-          
+
           IREmit->SetWriteCursor(CurrentIR.GetNode(Op->NewRIP));
 
           IREmit->ReplaceNodeArgument(CodeNode, 0, IREmit->_InlineConstant(Constant));
@@ -940,11 +951,11 @@ bool ConstProp::ConstantInlining(IREmitter *IREmit, const IRListView& CurrentIR)
         auto Op = IROp->CW<IR::IROp_LoadMem>();
 
         uint64_t Constant2{};
-        if (Op->OffsetType == MEM_OFFSET_SXTX && IREmit->IsValueConstant(Op->Header.Args[1], &Constant2)) {
+        if (Op->OffsetType == MEM_OFFSET_SXTX && IREmit->IsValueConstant(Op->Offset, &Constant2)) {
           if (IsImmMemory(Constant2, IROp->Size)) {
-            IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Header.Args[1]));
+            IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Offset));
 
-            IREmit->ReplaceNodeArgument(CodeNode, 1, IREmit->_InlineConstant(Constant2));
+            IREmit->ReplaceNodeArgument(CodeNode, Op->Offset_Index, IREmit->_InlineConstant(Constant2));
 
             Changed = true;
           }
@@ -957,11 +968,11 @@ bool ConstProp::ConstantInlining(IREmitter *IREmit, const IRListView& CurrentIR)
         auto Op = IROp->CW<IR::IROp_StoreMem>();
 
         uint64_t Constant2{};
-        if (Op->OffsetType == MEM_OFFSET_SXTX && IREmit->IsValueConstant(Op->Header.Args[2], &Constant2)) {
+        if (Op->OffsetType == MEM_OFFSET_SXTX && IREmit->IsValueConstant(Op->Offset, &Constant2)) {
           if (IsImmMemory(Constant2, IROp->Size)) {
-            IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Header.Args[2]));
+            IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Offset));
 
-            IREmit->ReplaceNodeArgument(CodeNode, 2, IREmit->_InlineConstant(Constant2));
+            IREmit->ReplaceNodeArgument(CodeNode, Op->Offset_Index, IREmit->_InlineConstant(Constant2));
 
             Changed = true;
           }

External/FEXCore/Source/Interface/IR/Passes/DeadCodeElimination.cpp

@@ -37,9 +37,27 @@ bool DeadCodeElimination::Run(IREmitter *IREmit) {
     while (1) {
       auto [CodeNode, IROp] = CodeLast();
 
+      bool HasSideEffects = IR::HasSideEffects(IROp->Op);
+      if (IROp->Op == OP_SYSCALL ||
+          IROp->Op == OP_INLINESYSCALL) {
+        FEXCore::IR::SyscallFlags Flags{};
+        if (IROp->Op == OP_SYSCALL) {
+          auto Op = IROp->C<IR::IROp_Syscall>();
+          Flags = Op->Flags;
+        }
+        else {
+          auto Op = IROp->C<IR::IROp_InlineSyscall>();
+          Flags = Op->Flags;
+        }
+
+        if ((Flags & FEXCore::IR::SyscallFlags::NOSIDEEFFECTS) == FEXCore::IR::SyscallFlags::NOSIDEEFFECTS) {
+          HasSideEffects = false;
+        }
+      }
+
       // Skip over anything that has side effects
       // Use count tracking can't safely remove anything with side effects
-      if (!IR::HasSideEffects(IROp->Op)) {
+      if (!HasSideEffects) {
         if (CodeNode->GetUses() == 0) {
           NumRemoved++;
           IREmit->Remove(CodeNode);

External/FEXCore/Source/Interface/IR/Passes/DeadContextStoreElimination.cpp

@@ -47,10 +47,12 @@ namespace {
     return (Type & ACCESS_TYPE_MASK) == ACCESS_READ;
   }
 
+  [[maybe_unused]]
   static bool IsInvalidAccess(LastAccessType Type) {
     return (Type & ACCESS_TYPE_MASK) == ACCESS_INVALID;
   }
 
+  [[maybe_unused]]
   static bool IsPartialAccess(LastAccessType Type) {
     return (Type & ACCESS_PARTIAL) == ACCESS_PARTIAL;
   }
@@ -254,7 +256,7 @@ namespace {
     });
 
 
-    size_t ClassifiedStructSize{};
+    [[maybe_unused]] size_t ClassifiedStructSize{};
     ContextClassificationInfo->Lookup.reserve(sizeof(FEXCore::Core::CPUState));
     for (auto &it : *ContextClassification) {
       LOGMAN_THROW_A_FMT(it.Class.Offset == ContextClassificationInfo->Lookup.size(), "Offset mismatch (offset={})", it.Class.Offset);
@@ -559,7 +561,7 @@ bool RCLSE::RedundantStoreLoadElimination(FEXCore::IR::IREmitter *IREmit) {
                 // the vector element
                 IREmit->SetWriteCursor(CodeNode);
                 // zext to size
-                LastNode = IREmit->_VMov(LastNode, IROp->Size);
+                LastNode = IREmit->_VMov(IROp->Size, LastNode);
 
                 IREmit->ReplaceAllUsesWithRange(CodeNode, LastNode, IREmit->GetIterator(IREmit->WrapNode(CodeNode)), BlockEnd);
                 RecordAccess(Info, Op->Class, Op->Offset, IROp->Size, ACCESS_READ, LastNode);
@@ -575,7 +577,7 @@ bool RCLSE::RedundantStoreLoadElimination(FEXCore::IR::IREmitter *IREmit) {
                        IROp->Size < IREmit->GetOpSize(LastNode)) {
               IREmit->SetWriteCursor(CodeNode);
               // trucate to size
-              LastNode = IREmit->_VMov(LastNode, IROp->Size);
+              LastNode = IREmit->_VMov(IROp->Size, LastNode);
               IREmit->ReplaceAllUsesWithRange(CodeNode, LastNode, IREmit->GetIterator(IREmit->WrapNode(CodeNode)), BlockEnd);
               RecordAccess(Info, Op->Class, Op->Offset, IROp->Size, ACCESS_READ, LastNode);
               Changed = true;
@@ -583,7 +585,7 @@ bool RCLSE::RedundantStoreLoadElimination(FEXCore::IR::IREmitter *IREmit) {
                        IROp->Size > IREmit->GetOpSize(LastNode)) {
               IREmit->SetWriteCursor(CodeNode);
               // zext to size
-              LastNode = IREmit->_VMov(LastNode, IROp->Size);
+              LastNode = IREmit->_VMov(IROp->Size, LastNode);
 
               IREmit->ReplaceAllUsesWithRange(CodeNode, LastNode, IREmit->GetIterator(IREmit->WrapNode(CodeNode)), BlockEnd);
               RecordAccess(Info, Op->Class, Op->Offset, IROp->Size, ACCESS_READ, LastNode);
@@ -659,10 +661,25 @@ bool RCLSE::RedundantStoreLoadElimination(FEXCore::IR::IREmitter *IREmit) {
           Changed = true;
         }
       }
+      else if (IROp->Op == OP_SYSCALL ||
+               IROp->Op == OP_INLINESYSCALL) {
+        FEXCore::IR::SyscallFlags Flags{};
+        if (IROp->Op == OP_SYSCALL) {
+          auto Op = IROp->C<IR::IROp_Syscall>();
+          Flags = Op->Flags;
+        }
+        else {
+          auto Op = IROp->C<IR::IROp_InlineSyscall>();
+          Flags = Op->Flags;
+        }
+
+        if ((Flags & FEXCore::IR::SyscallFlags::OPTIMIZETHROUGH) != FEXCore::IR::SyscallFlags::OPTIMIZETHROUGH) {
+          // We can't track through these
+          ResetClassificationAccesses(&LocalInfo);
+        }
+      }
       else if (IROp->Op == OP_STORECONTEXTINDEXED ||
                IROp->Op == OP_LOADCONTEXTINDEXED ||
-               IROp->Op == OP_SYSCALL ||
-               IROp->Op == OP_INLINESYSCALL ||
                IROp->Op == OP_BREAK) {
         // We can't track through these
         ResetClassificationAccesses(&LocalInfo);

External/FEXCore/Source/Interface/IR/Passes/DeadStoreElimination.cpp

@@ -77,7 +77,7 @@ struct FPRInfo {
 bool IsFPR(uint32_t Offset) {
 
   auto begin = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[0][0]);
-  auto end = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[17][0]);
+  auto end = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[16][0]);
 
   if (Offset < begin || Offset >= end)
     return false;

External/FEXCore/Source/Interface/IR/Passes/RegisterAllocationPass.cpp

@@ -15,6 +15,7 @@ $end_info$
 #include <FEXCore/Utils/BucketList.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXHeaderUtils/TypeDefines.h>
 
 #include <algorithm>
 #include <cstddef>
@@ -25,6 +26,7 @@ $end_info$
 #include <strings.h>
 #include <unordered_map>
 #include <unordered_set>
+#include <sys/user.h>
 #include <utility>
 #include <vector>
 
@@ -61,7 +63,7 @@ namespace {
   };
 
   static_assert(sizeof(RegisterNode) == 128 * 4);
-  constexpr size_t REGISTER_NODES_PER_PAGE = FEXCore::Core::PAGE_SIZE / sizeof(RegisterNode);
+  constexpr size_t REGISTER_NODES_PER_PAGE = FHU::FEX_PAGE_SIZE / sizeof(RegisterNode);
 
   struct RegisterSet {
     std::vector<RegisterClass> Classes;
@@ -570,10 +572,10 @@ namespace {
     // Get SRA Reg and Class from a Context offset
     auto GetRegAndClassFromOffset = [](uint32_t Offset) {
         auto beginGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[0]);
-        auto endGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[17]);
+        auto endGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[16]);
 
         auto beginFpr = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[0][0]);
-        auto endFpr = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[17][0]);
+        auto endFpr = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[16][0]);
 
         if (Offset >= beginGpr && Offset < endGpr) {
           auto reg = (Offset - beginGpr) / 8;
@@ -594,10 +596,10 @@ namespace {
     // Get a StaticMap entry from context offset
     const auto GetStaticMapFromOffset = [&](uint32_t Offset) -> LiveRange** {
         auto beginGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[0]);
-        auto endGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[17]);
+        auto endGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[16]);
 
         auto beginFpr = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[0][0]);
-        auto endFpr = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[17][0]);
+        auto endFpr = offsetof(FEXCore::Core::CpuStateFrame, State.xmm[16][0]);
 
         if (Offset >= beginGpr && Offset < endGpr) {
           auto reg = (Offset - beginGpr) / 8;
@@ -1115,8 +1117,6 @@ namespace {
         auto InterferenceNodeOpBeginIter = IR.at(InterferenceLiveRange->Begin);
         auto InterferenceNodeOpEndIter = IR.at(InterferenceLiveRange->End);
 
-        bool Found{};
-
         // If the nodes live range is entirely encompassed by the interference node's range
         // then spilling that range will /potentially/ lower RA
         // Will only lower register pressure if the interference node does NOT have a use inside of
@@ -1153,7 +1153,6 @@ namespace {
 
             const auto NextUseDistance = InterferenceNodeNextUse.ID().Value - CurrentLocation.Value;
             if (NextUseDistance >= InterferenceFarthestNextUse) {
-              Found = true;
               InterferenceIdToSpill = InterferenceNode;
               InterferenceFarthestNextUse = NextUseDistance;
             }
@@ -1501,7 +1500,7 @@ namespace {
       auto IROp = IR->GetNode(IR->GetNode(CodeBlock->Last)->Header.Previous)->Op(IR->GetData());
       if (IROp->Op == OP_JUMP) {
         auto Op = IROp->C<IROp_Jump>();
-        Graph->BlockPredecessors[Op->Target.ID()].insert(IR->GetID(BlockNode));
+        Graph->BlockPredecessors[Op->TargetBlock.ID()].insert(IR->GetID(BlockNode));
       } else if (IROp->Op == OP_CONDJUMP) {
         auto Op = IROp->C<IROp_CondJump>();
         Graph->BlockPredecessors[Op->TrueBlock.ID()].insert(IR->GetID(BlockNode));

External/FEXCore/Source/Interface/IR/Passes/StaticRegisterAllocationPass.cpp

@@ -25,7 +25,7 @@ public:
 
 bool IsStaticAllocGpr(uint32_t Offset, RegisterClassType Class) {
   const auto begin = offsetof(FEXCore::Core::CPUState, gregs[0]);
-  const auto end = offsetof(FEXCore::Core::CPUState, gregs[17]);
+  const auto end = offsetof(FEXCore::Core::CPUState, gregs[16]);
 
   if (Offset >= begin && Offset < end) {
     const auto reg = (Offset - begin) / 8;
@@ -40,7 +40,7 @@ bool IsStaticAllocGpr(uint32_t Offset, RegisterClassType Class) {
 
 bool IsStaticAllocFpr(uint32_t Offset, RegisterClassType Class, bool AllowGpr) {
   const auto begin = offsetof(FEXCore::Core::CPUState, xmm[0][0]);
-  const auto end = offsetof(FEXCore::Core::CPUState, xmm[17][0]);
+  const auto end = offsetof(FEXCore::Core::CPUState, xmm[16][0]);
 
   if (Offset >= begin && Offset < end) {
     const auto reg = (Offset - begin) / 16;

External/FEXCore/Source/Interface/IR/Passes/SyscallOptimization.cpp

@@ -26,14 +26,20 @@ bool SyscallOptimization::Run(IREmitter *IREmit) {
   bool Changed = false;
   auto CurrentIR = IREmit->ViewIR();
 
-
   for (auto [CodeNode, IROp] : CurrentIR.GetAllCode()) {
 
     if (IROp->Op == FEXCore::IR::OP_SYSCALL) {
+      auto Op = IROp->CW<IR::IROp_Syscall>();
+
       // Is the first argument a constant?
       uint64_t Constant;
-      if (IREmit->IsValueConstant(IROp->Args[0], &Constant)) {
+      if (IREmit->IsValueConstant(Op->SyscallID, &Constant)) {
         auto SyscallDef = Manager->SyscallHandler->GetSyscallABI(Constant);
+        auto SyscallFlags = Manager->SyscallHandler->GetSyscallFlags(Constant);
+
+        // Update the syscall flags
+        Op->Flags = SyscallFlags;
+
         // XXX: Once we have the ability to do real function calls then we can call directly in to the syscall handler
         if (SyscallDef.NumArgs < FEXCore::HLE::SyscallArguments::MAX_ARGS) {
           // If the number of args are less than what the IR op supports then we can remove arg usage
@@ -53,7 +59,8 @@ bool SyscallOptimization::Run(IREmitter *IREmit) {
               CurrentIR.GetNode(IROp->Args[4]),
               CurrentIR.GetNode(IROp->Args[5]),
               CurrentIR.GetNode(IROp->Args[6]),
-              SyscallDef.HostSyscallNumber);
+              SyscallDef.HostSyscallNumber,
+              Op->Flags);
 
             // Replace all syscall uses with this inline one
             IREmit->ReplaceAllUsesWith(CodeNode, InlineSyscall);
@@ -62,8 +69,9 @@ bool SyscallOptimization::Run(IREmitter *IREmit) {
             IREmit->Remove(CodeNode);
           }
 #endif
-          Changed = true;
         }
+
+        Changed = true;
       }
     }
   }

External/FEXCore/Source/Utils/Allocator.cpp

@@ -3,9 +3,11 @@
 #include <FEXCore/Utils/CompilerDefs.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXHeaderUtils/Syscalls.h>
+#include <FEXHeaderUtils/TypeDefines.h>
 
 #include <array>
 #include <sys/mman.h>
+#include <sys/user.h>
 #ifdef ENABLE_JEMALLOC
 #include <jemalloc/jemalloc.h>
 #endif
@@ -109,10 +111,10 @@ namespace FEXCore::Allocator {
         for (int i = 0; i < 64; ++i) {
           // Try grabbing a some of the top pages of the range
           // x86 allocates some high pages in the top end
-          void *Ptr = ::mmap(reinterpret_cast<void*>(Size - PAGE_SIZE * i), PAGE_SIZE, PROT_NONE, MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+          void *Ptr = ::mmap(reinterpret_cast<void*>(Size - FHU::FEX_PAGE_SIZE * i), FHU::FEX_PAGE_SIZE, PROT_NONE, MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
           if (Ptr != (void*)~0ULL) {
-            ::munmap(Ptr, PAGE_SIZE);
-            if (Ptr == (void*)(Size - PAGE_SIZE * i)) {
+            ::munmap(Ptr, FHU::FEX_PAGE_SIZE);
+            if (Ptr == (void*)(Size - FHU::FEX_PAGE_SIZE * i)) {
               return true;
             }
           }

External/FEXCore/Source/Utils/Allocator/64BitAllocator.cpp

@@ -4,7 +4,9 @@
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXHeaderUtils/ScopedSignalMask.h>
 #include <FEXHeaderUtils/Syscalls.h>
+#include <FEXHeaderUtils/TypeDefines.h>
 
 #include <algorithm>
 #include <array>
@@ -19,13 +21,12 @@
 #include <sstream>
 #include <sys/mman.h>
 #include <sys/utsname.h>
+#include <sys/user.h>
 #include <type_traits>
 #include <utility>
 
-static constexpr uint64_t PAGE_SHIFT = 12;
-static constexpr uint64_t PAGE_MASK = (1 << PAGE_SHIFT) - 1;
-
 namespace Alloc::OSAllocator {
+
   class OSAllocator_64Bit final : public Alloc::HostAllocator {
     public:
       OSAllocator_64Bit();
@@ -37,15 +38,14 @@ namespace Alloc::OSAllocator {
       int Munmap(void *addr, size_t length) override;
 
     private:
-      constexpr static uint64_t PAGE_SIZE = 4096;
       // Upper bound is the maximum virtual address space of the host processor
       uintptr_t UPPER_BOUND = (1ULL << 57);
 
       // Lower bound is the starting of the range just past the lower 32bits
       constexpr static uintptr_t LOWER_BOUND = 0x1'0000'0000ULL;
 
-      uintptr_t UPPER_BOUND_PAGE = UPPER_BOUND / PAGE_SIZE;
-      constexpr static uintptr_t LOWER_BOUND_PAGE = LOWER_BOUND / PAGE_SIZE;
+      uintptr_t UPPER_BOUND_PAGE = UPPER_BOUND / FHU::FEX_PAGE_SIZE;
+      constexpr static uintptr_t LOWER_BOUND_PAGE = LOWER_BOUND / FHU::FEX_PAGE_SIZE;
 
       struct ReservedVMARegion {
         uintptr_t Base;
@@ -82,19 +82,19 @@ namespace Alloc::OSAllocator {
           // 0x100'0000 Pages
           // 1 bit per page for tracking means 0x20'0000 (Pages / 8) bytes of flex space
           // Which is 2MB of tracking
-          uint64_t NumElements = (Size >> PAGE_SHIFT) * sizeof(uint64_t);
+          uint64_t NumElements = (Size >> FHU::FEX_PAGE_SHIFT) * sizeof(uint64_t);
           return sizeof(LiveVMARegion) + FEXCore::FlexBitSet<uint64_t>::Size(NumElements);
         }
 
         static void InitializeVMARegionUsed(LiveVMARegion *Region, size_t AdditionalSize) {
-          size_t SizeOfLiveRegion = FEXCore::AlignUp(LiveVMARegion::GetSizeWithFlexSet(Region->SlabInfo->RegionSize), PAGE_SIZE);
+          size_t SizeOfLiveRegion = FEXCore::AlignUp(LiveVMARegion::GetSizeWithFlexSet(Region->SlabInfo->RegionSize), FHU::FEX_PAGE_SIZE);
           size_t SizePlusManagedData = SizeOfLiveRegion + AdditionalSize;
 
           Region->FreeSpace = Region->SlabInfo->RegionSize - SizePlusManagedData;
 
-          size_t NumPages = SizePlusManagedData >> PAGE_SHIFT;
+          size_t NumPages = SizePlusManagedData >> FHU::FEX_PAGE_SHIFT;
           // Memset the full tracking to zero to state nothing used
-          Region->UsedPages.MemSet(Region->SlabInfo->RegionSize >> PAGE_SHIFT);
+          Region->UsedPages.MemSet(Region->SlabInfo->RegionSize >> FHU::FEX_PAGE_SHIFT);
           // Set our reserved pages
           for (size_t i = 0; i < NumPages; ++i) {
             // Set our used pages
@@ -106,8 +106,8 @@ namespace Alloc::OSAllocator {
       static_assert(std::is_trivially_copyable<LiveVMARegion>::value, "Needs to be trivially copyable");
       static_assert(offsetof(LiveVMARegion, UsedPages) == sizeof(LiveVMARegion), "FlexBitSet needs to be at the end");
 
-      using ReservedRegionListType = std::pmr::list<ReservedVMARegion*>;
-      using LiveRegionListType = std::pmr::list<LiveVMARegion*>;
+      using ReservedRegionListType = fex_pmr::list<ReservedVMARegion*>;
+      using LiveRegionListType = fex_pmr::list<LiveVMARegion*>;
       ReservedRegionListType *ReservedRegions{};
       LiveRegionListType *LiveRegions{};
 
@@ -120,7 +120,7 @@ namespace Alloc::OSAllocator {
 
         ReservedRegions->erase(ReservedIterator);
         // mprotect the new region we've allocated
-        size_t SizeOfLiveRegion = FEXCore::AlignUp(LiveVMARegion::GetSizeWithFlexSet(ReservedRegion->RegionSize), PAGE_SIZE);
+        size_t SizeOfLiveRegion = FEXCore::AlignUp(LiveVMARegion::GetSizeWithFlexSet(ReservedRegion->RegionSize), FHU::FEX_PAGE_SIZE);
         size_t SizePlusManagedData = UsedSize + SizeOfLiveRegion;
 
         [[maybe_unused]] auto Res = mprotect(reinterpret_cast<void*>(ReservedRegion->Base), SizePlusManagedData, PROT_READ | PROT_WRITE);
@@ -148,7 +148,7 @@ void OSAllocator_64Bit::DetermineVASize() {
   size_t Bits = FEXCore::Allocator::DetermineVASize();
   uintptr_t Size = 1ULL << Bits;
   UPPER_BOUND = Size;
-  UPPER_BOUND_PAGE = UPPER_BOUND / PAGE_SIZE;
+  UPPER_BOUND_PAGE = UPPER_BOUND / FHU::FEX_PAGE_SIZE;
 }
 
 void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) {
@@ -161,13 +161,13 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
 
   uint64_t Addr = reinterpret_cast<uint64_t>(addr);
   // Addr must be page aligned
-  if (Addr & PAGE_MASK) {
+  if (Addr & ~FHU::FEX_PAGE_MASK) {
     return reinterpret_cast<void*>(-EINVAL);
   }
 
   // If FD is provided then offset must also be page aligned
   if (fd != -1 &&
-      offset & PAGE_MASK) {
+      offset & ~FHU::FEX_PAGE_MASK) {
     return reinterpret_cast<void*>(-EINVAL);
   }
 
@@ -177,13 +177,13 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
   }
 
   bool Fixed = (flags & MAP_FIXED) || (flags & MAP_FIXED_NOREPLACE);
-  length = FEXCore::AlignUp(length, PAGE_SIZE);
+  length = FEXCore::AlignUp(length, FHU::FEX_PAGE_SIZE);
 
   uint64_t AddrEnd = Addr + length;
-  size_t NumberOfPages = length / PAGE_SIZE;
+  size_t NumberOfPages = length / FHU::FEX_PAGE_SIZE;
 
   // This needs a mutex to be thread safe
-  std::scoped_lock<std::mutex> lk{AllocationMutex};
+  FHU::ScopedSignalMaskWithMutex lk(AllocationMutex);
 
   uint64_t AllocatedOffset{};
   LiveVMARegion *LiveRegion{};
@@ -224,14 +224,14 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
 
   auto CheckIfRangeFits = [&AllocatedOffset](LiveVMARegion *Region, uint64_t length, int prot, int flags, int fd, off_t offset, uint64_t StartingPosition = 0) -> std::pair<LiveVMARegion*, void*> {
     uint64_t AllocatedPage{};
-    uint64_t NumberOfPages = length >> PAGE_SHIFT;
+    uint64_t NumberOfPages = length >> FHU::FEX_PAGE_SHIFT;
 
     if (Region->FreeSpace >= length) {
       uint64_t LastAllocation =
         StartingPosition ?
-        (StartingPosition - Region->SlabInfo->Base) >> PAGE_SHIFT
+        (StartingPosition - Region->SlabInfo->Base) >> FHU::FEX_PAGE_SHIFT
         : Region->LastPageAllocation;
-      size_t RegionNumberOfPages = Region->SlabInfo->RegionSize >> PAGE_SHIFT;
+      size_t RegionNumberOfPages = Region->SlabInfo->RegionSize >> FHU::FEX_PAGE_SHIFT;
 
       // Backward scan
       // We need to do a backward scan first to fill any holes
@@ -299,7 +299,7 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
       }
 
       if (AllocatedPage) {
-        AllocatedOffset = Region->SlabInfo->Base + AllocatedPage * PAGE_SIZE;
+        AllocatedOffset = Region->SlabInfo->Base + AllocatedPage * FHU::FEX_PAGE_SIZE;
 
         // We need to setup protections for this
         void *MMapResult = ::mmap(reinterpret_cast<void*>(AllocatedOffset),
@@ -389,7 +389,7 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
     if (!LiveRegion) {
       // Couldn't find a fit in the live regions
       // Allocate a new reserved region
-      size_t lengthOfLiveRegion = FEXCore::AlignUp(LiveVMARegion::GetSizeWithFlexSet(length), PAGE_SIZE);
+      size_t lengthOfLiveRegion = FEXCore::AlignUp(LiveVMARegion::GetSizeWithFlexSet(length), FHU::FEX_PAGE_SIZE);
       size_t lengthPlusManagedData = length + lengthOfLiveRegion;
       for (auto it = ReservedRegions->begin(); it != ReservedRegions->end(); ++it) {
         if ((*it)->RegionSize >= lengthPlusManagedData) {
@@ -403,7 +403,7 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
   if (LiveRegion) {
     // Mark the pages as used
     uintptr_t RegionBegin = LiveRegion->SlabInfo->Base;
-    uintptr_t MappedBegin = (AllocatedOffset - RegionBegin) >> PAGE_SHIFT;
+    uintptr_t MappedBegin = (AllocatedOffset - RegionBegin) >> FHU::FEX_PAGE_SHIFT;
 
     for (size_t i = 0; i < NumberOfPages; ++i) {
       LiveRegion->UsedPages.Set(MappedBegin + i);
@@ -429,11 +429,11 @@ int OSAllocator_64Bit::Munmap(void *addr, size_t length) {
 
   uint64_t Addr = reinterpret_cast<uint64_t>(addr);
 
-  if (Addr & PAGE_MASK) {
+  if (Addr & ~FHU::FEX_PAGE_MASK) {
     return -EINVAL;
   }
 
-  if (length & PAGE_MASK) {
+  if (length & ~FHU::FEX_PAGE_MASK) {
     return -EINVAL;
   }
 
@@ -442,9 +442,9 @@ int OSAllocator_64Bit::Munmap(void *addr, size_t length) {
   }
 
   // This needs a mutex to be thread safe
-  std::scoped_lock<std::mutex> lk{AllocationMutex};
+  FHU::ScopedSignalMaskWithMutex lk(AllocationMutex);
 
-  length = FEXCore::AlignUp(length, PAGE_SIZE);
+  length = FEXCore::AlignUp(length, FHU::FEX_PAGE_SIZE);
 
   uintptr_t PtrBegin = reinterpret_cast<uintptr_t>(addr);
   uintptr_t PtrEnd = PtrBegin + length;
@@ -458,8 +458,8 @@ int OSAllocator_64Bit::Munmap(void *addr, size_t length) {
       // Live region fully encompasses slab range
 
       uint64_t FreedPages{};
-      uint32_t SlabPageBegin = (PtrBegin - RegionBegin) >> PAGE_SHIFT;
-      uint64_t PagesToFree = length >> PAGE_SHIFT;
+      uint32_t SlabPageBegin = (PtrBegin - RegionBegin) >> FHU::FEX_PAGE_SHIFT;
+      uint64_t PagesToFree = length >> FHU::FEX_PAGE_SHIFT;
 
       for (size_t i = 0; i < PagesToFree; ++i) {
         FreedPages += (*it)->UsedPages.TestAndClear(SlabPageBegin + i) ? 1 : 0;
@@ -621,8 +621,8 @@ OSAllocator_64Bit::OSAllocator_64Bit() {
 }
 
 OSAllocator_64Bit::~OSAllocator_64Bit() {
-  // For consistency, pull the mutex
-  std::scoped_lock<std::mutex> lk{AllocationMutex};
+  // This needs a mutex to be thread safe
+  FHU::ScopedSignalMaskWithMutex lk(AllocationMutex);
 
   // Walk the pages and deallocate
   // First walk the live regions

External/FEXCore/Source/Utils/Allocator/HostAllocator.h

@@ -5,8 +5,6 @@
 #include <memory>
 #include <sys/types.h>
 
-constexpr static uint64_t PAGE_SIZE = 4096;
-
 namespace Alloc {
   // HostAllocator is just a page pased slab allocator
   // Similar to mmap and munmap only mapping at the page level

External/FEXCore/Source/Utils/Allocator/IntrusiveArenaAllocator.h

@@ -4,15 +4,30 @@
 #include "HostAllocator.h"
 
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXHeaderUtils/TypeDefines.h>
 
 #include <bitset>
 #include <cstddef>
+#ifdef TERMUX_BUILD
+#ifdef __has_include
+#if __has_include(<memory_resource>)
+#error Termux <experimental/memory_resource> workaround can be removed
+#endif
+#endif
+#include <experimental/memory_resource>
+#include <experimental/list>
+namespace fex_pmr = std::experimental::pmr;
+#else
 #include <memory_resource>
+namespace fex_pmr = std::pmr;
+#endif
+#include <sys/user.h>
+
 #include <mutex>
 #include <vector>
 
 namespace Alloc {
-  class ForwardOnlyIntrusiveArenaAllocator final : public std::pmr::memory_resource {
+  class ForwardOnlyIntrusiveArenaAllocator final : public fex_pmr::memory_resource {
   public:
     ForwardOnlyIntrusiveArenaAllocator(void* Ptr, size_t _Size)
       : Begin {reinterpret_cast<uintptr_t>(Ptr)}
@@ -54,7 +69,7 @@ namespace Alloc {
       // Do nothing
     }
 
-    bool do_is_equal(const std::pmr::memory_resource& other) const noexcept override {
+    bool do_is_equal(const fex_pmr::memory_resource& other) const noexcept override {
       // Only if the allocator pointers are the same are they equal
       if (this == &other) {
         return true;
@@ -68,14 +83,14 @@ namespace Alloc {
     size_t LastAllocation{};
   };
 
-  class IntrusiveArenaAllocator final : public std::pmr::memory_resource {
+  class IntrusiveArenaAllocator final : public fex_pmr::memory_resource {
   public:
     IntrusiveArenaAllocator(void* Ptr, size_t _Size)
       : Begin {reinterpret_cast<uintptr_t>(Ptr)}
       , Size {_Size} {
-      uint64_t NumberOfPages = _Size / PAGE_SIZE;
+      uint64_t NumberOfPages = _Size / FHU::FEX_PAGE_SIZE;
       uint64_t UsedBits = FEXCore::AlignUp(sizeof(IntrusiveArenaAllocator) +
-        Size / PAGE_SIZE / 8, PAGE_SIZE);
+        Size / FHU::FEX_PAGE_SIZE / 8, FHU::FEX_PAGE_SIZE);
       for (size_t i = 0; i < UsedBits; ++i) {
         UsedPages.Set(i);
       }
@@ -103,7 +118,7 @@ namespace Alloc {
     void *do_allocate(std::size_t bytes, std::size_t alignment) override {
       std::scoped_lock<std::mutex> lk{AllocationMutex};
 
-      size_t NumberPages = FEXCore::AlignUp(bytes, PAGE_SIZE) / PAGE_SIZE;
+      size_t NumberPages = FEXCore::AlignUp(bytes, FHU::FEX_PAGE_SIZE) / FHU::FEX_PAGE_SIZE;
 
       uintptr_t AllocatedOffset{};
 
@@ -147,7 +162,7 @@ namespace Alloc {
         LastAllocatedPageOffset = AllocatedOffset + NumberPages;
 
         // Now convert this base page to a pointer and return it
-        return reinterpret_cast<void*>(Begin + AllocatedOffset * PAGE_SIZE);
+        return reinterpret_cast<void*>(Begin + AllocatedOffset * FHU::FEX_PAGE_SIZE);
       }
 
       return nullptr;
@@ -156,8 +171,8 @@ namespace Alloc {
     void do_deallocate(void* p, std::size_t bytes, std::size_t alignment) override {
       std::scoped_lock<std::mutex> lk{AllocationMutex};
 
-      uintptr_t PageOffset = (reinterpret_cast<uintptr_t>(p) - Begin) / PAGE_SIZE;
-      size_t NumPages = FEXCore::AlignUp(bytes, PAGE_SIZE) / PAGE_SIZE;
+      uintptr_t PageOffset = (reinterpret_cast<uintptr_t>(p) - Begin) / FHU::FEX_PAGE_SIZE;
+      size_t NumPages = FEXCore::AlignUp(bytes, FHU::FEX_PAGE_SIZE) / FHU::FEX_PAGE_SIZE;
 
       // Walk the allocation list and deallocate
       uint64_t FreedPages{};
@@ -167,7 +182,7 @@ namespace Alloc {
       FreePages += FreedPages;
     }
 
-    bool do_is_equal(const std::pmr::memory_resource& other) const noexcept override {
+    bool do_is_equal(const fex_pmr::memory_resource& other) const noexcept override {
       // Only if the allocator pointers are the same are they equal
       if (this == &other) {
         return true;

External/FEXCore/Source/Utils/MemberFunctionToPointer.h

@@ -0,0 +1,50 @@
+#pragma once
+
+#include <FEXCore/Utils/LogManager.h>
+
+#include <cstdint>
+
+namespace FEXCore::Utils {
+
+/**
+ * @brief Casts a class's member function pointer to a raw pointer that we can JIT
+ *
+ * Has additional validation to ensure we aren't casting a class member that is invalid
+ */
+template <typename PointerToMemberType>
+class MemberFunctionToPointerCast final {
+  public:
+    MemberFunctionToPointerCast(PointerToMemberType Function) {
+      memcpy(&PMF, &Function, sizeof(PMF));
+
+#ifdef _M_X86_64
+      // Itanium C++ ABI (https://itanium-cxx-abi.github.io/cxx-abi/abi.html#member-function-pointers)
+      // Low bit of ptr specifies if this Member function pointer is virtual or not
+      // Throw an assert if we were trying to cast a virtual member
+      LOGMAN_THROW_A_FMT((PMF.ptr & 1) == 0, "C++ Pointer-To-Member representation didn't have low bit set to 0. Are you trying to cast a virtual member?");
+#elif defined(_M_ARM_64 )
+      // C++ ABI for the Arm 64-bit Architecture (IHI 0059E)
+      // 4.2.1 Representation of pointer to member function
+      // Differs from Itanium specification
+      LOGMAN_THROW_A_FMT(PMF.adj == 0, "C++ Pointer-To-Member representation didn't have adj == 0. Are you trying to cast a virtual member?");
+#else
+#error Don't know how to cast Member to function here. Likely just Itanium
+#endif
+    }
+
+    uintptr_t GetConvertedPointer() const {
+      return PMF.ptr;
+    }
+
+  private:
+    struct PointerToMember {
+      uintptr_t ptr;
+      uintptr_t adj;
+    };
+
+    PointerToMember PMF;
+
+    // Ensure the representation of PointerToMember matches
+    static_assert(sizeof(PMF) == sizeof(PointerToMemberType));
+};
+}

External/FEXCore/Source/Utils/Telemetry.cpp

@@ -15,9 +15,13 @@ namespace FEXCore::Telemetry {
   static std::array<Value, FEXCore::Telemetry::TelemetryType::TYPE_LAST> TelemetryValues = {{ }};
   const std::array<std::string_view, FEXCore::Telemetry::TelemetryType::TYPE_LAST> TelemetryNames {
     "64byte Split Locks",
-    "16Byte Split atomics",
+    "16byte Split atomics",
     "VEX instructions (AVX)",
     "EVEX instructions (AVX512)",
+    "16bit CAS Tear",
+    "32bit CAS Tear",
+    "64bit CAS Tear",
+    "128bit CAS Tear",
   };
   void Initialize() {
     auto DataDirectory = Config::GetDataDirectory();

External/FEXCore/Source/Utils/Threads.cpp

@@ -31,7 +31,7 @@ namespace FEXCore::Threads {
     std::lock_guard lk{DeadStackPoolMutex};
     if (DeadStackPool.size() == 0) {
       // Nothing in the pool, just allocate
-      return FEXCore::Allocator::mmap(nullptr, Size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN, -1, 0);
+      return FEXCore::Allocator::mmap(nullptr, Size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
     }
 
     // Keep the first item in the stack pool

External/FEXCore/include/FEXCore/Core/CPUBackend.h

@@ -92,6 +92,13 @@ class LLVMCore;
     virtual void CopyNecessaryDataForCompileThread(CPUBackend *Original) {}
     virtual bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true, bool IncludeCompileService = true) const { return false; }
 
+    /**
+     * @brief Does this CPUBackend need its IR to stick around for correct emulation
+     *
+     * This should only be used on the interpreter, all other backends can clear their IR
+     */
+    virtual bool NeedsRetainedIRCopy() const { return false; }
+
     using AsmDispatch = FEX_NAKED void(*)(FEXCore::Core::CpuStateFrame *Frame);
     using JITCallback = FEX_NAKED void(*)(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP);
 

External/FEXCore/include/FEXCore/Core/CodeLoader.h

@@ -48,6 +48,8 @@ public:
 
   using IRHandler = std::function<void(uint64_t Addr, FEXCore::IR::IREmitter *IR)>;
   virtual void AddIR(IRHandler Handler) {}
+
+  virtual uint64_t GetBaseOffset() const { return 0; }
 };
 
 

External/FEXCore/include/FEXCore/Core/CoreState.h

@@ -32,6 +32,124 @@ namespace FEXCore::Core {
 
   struct InternalThreadState;
 
+  enum FallbackHandlerIndex {
+    OPINDEX_F80LOADFCW = 0,
+    OPINDEX_F80CVTTO_4,
+    OPINDEX_F80CVTTO_8,
+    OPINDEX_F80CVT_4,
+    OPINDEX_F80CVT_8,
+    OPINDEX_F80CVTINT_2,
+    OPINDEX_F80CVTINT_4,
+    OPINDEX_F80CVTINT_8,
+    OPINDEX_F80CVTINT_TRUNC2,
+    OPINDEX_F80CVTINT_TRUNC4,
+    OPINDEX_F80CVTINT_TRUNC8,
+    OPINDEX_F80CMP_0,
+    OPINDEX_F80CMP_1,
+    OPINDEX_F80CMP_2,
+    OPINDEX_F80CMP_3,
+    OPINDEX_F80CMP_4,
+    OPINDEX_F80CMP_5,
+    OPINDEX_F80CMP_6,
+    OPINDEX_F80CMP_7,
+    OPINDEX_F80CVTTOINT_2,
+    OPINDEX_F80CVTTOINT_4,
+
+    // Unary
+    OPINDEX_F80ROUND,
+    OPINDEX_F80F2XM1,
+    OPINDEX_F80TAN,
+    OPINDEX_F80SQRT,
+    OPINDEX_F80SIN,
+    OPINDEX_F80COS,
+    OPINDEX_F80XTRACT_EXP,
+    OPINDEX_F80XTRACT_SIG,
+    OPINDEX_F80BCDSTORE,
+    OPINDEX_F80BCDLOAD,
+
+    // Binary
+    OPINDEX_F80ADD,
+    OPINDEX_F80SUB,
+    OPINDEX_F80MUL,
+    OPINDEX_F80DIV,
+    OPINDEX_F80FYL2X,
+    OPINDEX_F80ATAN,
+    OPINDEX_F80FPREM1,
+    OPINDEX_F80FPREM,
+    OPINDEX_F80SCALE,
+
+    // Maximum
+    OPINDEX_MAX,
+  };
+
+  union JITPointers {
+    struct {
+      // Process specific
+      uint64_t LUDIV{};
+      uint64_t LDIV{};
+      uint64_t LUREM{};
+      uint64_t LREM{};
+      uint64_t PrintValue{};
+      uint64_t PrintVectorValue{};
+      uint64_t RemoveCodeEntryFromJIT{};
+      uint64_t CPUIDObj{};
+      uint64_t CPUIDFunction{};
+      uint64_t SyscallHandlerObj{};
+      uint64_t SyscallHandlerFunc{};
+
+      uint64_t FallbackHandlerPointers[FallbackHandlerIndex::OPINDEX_MAX];
+
+      // Thread Specific
+      uint64_t SignalHandlerRefCountPointer{};
+
+      /**
+       * @name Dispatcher pointers
+       * @{ */
+      uint64_t DispatcherLoopTop{};
+      uint64_t DispatcherLoopTopFillSRA{};
+      uint64_t ThreadStopHandlerSpillSRA{};
+      uint64_t ThreadPauseHandlerSpillSRA{};
+      uint64_t UnimplementedInstructionHandler{};
+      uint64_t OverflowExceptionHandler{};
+      uint64_t SignalReturnHandler{};
+      uint64_t L1Pointer{};
+      uint64_t LUDIVHandler{};
+      uint64_t LDIVHandler{};
+      uint64_t LUREMHandler{};
+      uint64_t LREMHandler{};
+      /**  @} */
+    } AArch64;
+
+    struct {
+      // Process specific
+      uint64_t PrintValue{};
+      uint64_t PrintVectorValue{};
+      uint64_t RemoveCodeEntryFromJIT{};
+      uint64_t CPUIDObj{};
+      uint64_t CPUIDFunction{};
+      uint64_t SyscallHandlerObj{};
+      uint64_t SyscallHandlerFunc{};
+
+      uint64_t FallbackHandlerPointers[FallbackHandlerIndex::OPINDEX_MAX];
+
+      // Thread Specific
+      uint64_t SignalHandlerRefCountPointer{};
+
+      /**
+       * @name Dispatcher pointers
+       * @{ */
+      uint64_t DispatcherLoopTop{};
+      uint64_t DispatcherLoopTopFillSRA{};
+      uint64_t ThreadStopHandler{};
+      uint64_t ThreadPauseHandler{};
+      uint64_t UnimplementedInstructionHandler{};
+      uint64_t OverflowExceptionHandler{};
+      uint64_t SignalReturnHandler{};
+      uint64_t L1Pointer{};
+      /**  @} */
+    } X86;
+  };
+
   // Each guest JIT frame has one of these
   struct CpuStateFrame {
     CPUState State;
@@ -52,18 +170,17 @@ namespace FEXCore::Core {
      */
     uint64_t InSyscallInfo{};
     InternalThreadState* Thread;
+
+    // Pointers that the JIT needs to load to remove relocations
+    JITPointers Pointers;
   };
   static_assert(offsetof(CpuStateFrame, State) == 0, "CPUState must be first member in CpuStateFrame");
   static_assert(offsetof(CpuStateFrame, State.rip) == 0, "rip must be zero offset in CpuStateFrame");
+  static_assert(offsetof(CpuStateFrame, Pointers) % 8 == 0, "JITPointers need to be aligned to 8 bytes");
+  static_assert(offsetof(CpuStateFrame, Pointers) + sizeof(CpuStateFrame::Pointers) <= 32760, "JITPointers maximum pointer needs to be less than architecture maximum 32768");
+
   static_assert(std::is_standard_layout<CpuStateFrame>::value, "This needs to be standard layout");
 
-#ifdef PAGE_SIZE
-  static_assert(PAGE_SIZE == 4096, "FEX only supports 4k pages");
-#undef PAGE_SIZE
-#endif
-
-  constexpr uint64_t PAGE_SIZE = 4096;
-
   FEX_DEFAULT_VISIBILITY std::string_view const& GetFlagName(unsigned Flag);
   FEX_DEFAULT_VISIBILITY std::string_view const& GetGRegName(unsigned Reg);
 }

External/FEXCore/include/FEXCore/Core/SignalDelegator.h

@@ -3,12 +3,12 @@
 #include <FEXCore/Utils/CompilerDefs.h>
 
 #include <array>
-#include <bits/types/siginfo_t.h>
-#include <bits/types/stack_t.h>
 #include <cstdint>
 #include <functional>
 #include <utility>
+#include <signal.h>
 #include <stddef.h>
+#include <vector>
 
 namespace FEXCore {
 namespace Core {
@@ -97,14 +97,14 @@ namespace Core {
 
   private:
     struct HostSignalHandler {
-      FEXCore::HostSignalDelegatorFunction Handler{};
+      std::vector<FEXCore::HostSignalDelegatorFunction> Handlers{};
       FEXCore::HostSignalDelegatorFunction FrontendHandler{};
     };
     std::array<HostSignalHandler, MAX_SIGNALS + 1> HostHandlers{};
 
   protected:
     void SetHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
-      HostHandlers[Signal].Handler = std::move(Func);
+      HostHandlers[Signal].Handlers.push_back(std::move(Func));
     }
     void SetFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
       HostHandlers[Signal].FrontendHandler = std::move(Func);

External/FEXCore/include/FEXCore/Core/UContext.h

@@ -155,23 +155,59 @@ namespace FEXCore {
         } _timer;
       } _sifields;
 
+      union HostSigInfo_t {
+        // This anonymous struct needs to match the host definition
+        struct {
+          uint32_t si_signo;
+          uint32_t si_errno;
+          uint32_t si_code;
+
+          uint32_t __pad0;
+
+          // Pad[28] is a union for all the sifields
+          uint32_t _pad[28];
+        } FEXDef;
+        ::siginfo_t host{};
+      };
+      static_assert(sizeof(HostSigInfo_t) == 128, "This needs to be the right size");
+
       siginfo_t() = delete;
 
       operator ::siginfo_t() const {
-        ::siginfo_t val{};
-        val.si_signo = si_signo;
-        val.si_errno = si_errno;
-        val.si_code = si_code;
+        // The definition of siginfo_t changes depending on the host environment
+        // It is guaranteed to be 128 bytes and the kernel interface is the same for all of them
+        // Since we only run on Linux
+        HostSigInfo_t val{};
+
+        val.FEXDef.si_signo = si_signo;
+        val.FEXDef.si_errno = si_errno;
+        val.FEXDef.si_code = si_code;
 
         // Host siginfo has a pad member that is set to zeros
-        val.__pad0 = 0;
+        val.FEXDef.__pad0 = 0;
 
         // Copy over the union
         // The union is different sizes on 64-bit versus 32-bit
-        memcpy(val._sifields._pad, _sifields.pad, std::min(sizeof(val._sifields._pad), sizeof(_sifields.pad)));
+        memcpy(val.FEXDef._pad, _sifields.pad, std::min(sizeof(val.FEXDef._pad), sizeof(_sifields.pad)));
 
-        return val;
+        return val.host;
       }
+
+      siginfo_t(::siginfo_t val) {
+        HostSigInfo_t host;
+        host.host = val;
+
+        si_signo = host.FEXDef.si_signo;
+        si_errno = host.FEXDef.si_errno;
+        si_code = host.FEXDef.si_code;
+
+        // Copy over the union
+        // The union is different sizes on 64-bit versus 32-bit
+        memcpy(_sifields.pad, host.FEXDef._pad, std::min(sizeof(host.FEXDef._pad), sizeof(_sifields.pad)));
+      }
+      static_assert(offsetof(::siginfo_t, si_signo) == offsetof(HostSigInfo_t, FEXDef.si_signo), "si_signo in wrong location?");
+      static_assert(offsetof(::siginfo_t, si_errno) == offsetof(HostSigInfo_t, FEXDef.si_errno), "si_errno in wrong location?");
+      static_assert(offsetof(::siginfo_t, si_code) == offsetof(HostSigInfo_t, FEXDef.si_code), "si_code in wrong location?");
     };
     static_assert(sizeof(FEXCore::x86::siginfo_t) == 128, "This needs to be the right size");