Docs: Update for release FEX-2512

SyscallsSMCTracking: Workaround assert in ELF mapping

.clang-format-ignore

@@ -7,3 +7,6 @@ FEXCore/Source/Interface/Core/X86Tables/*
 # Inline headers with list-like content that can't be processed individually
 Source/Tools/LinuxEmulation/LinuxSyscalls/x*/SyscallsNames.inl
 Source/Tools/LinuxEmulation/LinuxSyscalls/x*/Ioctl/*.inl
+
+# Include files in unittests
+unittests/*ASM/Includes/*.inc

.github/workflows/steamrt4.yml

@@ -0,0 +1,79 @@
+name: steamrt4 build
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+env:
+  BUILD_TYPE: Release
+  CC: clang
+  CXX: clang++
+
+jobs:
+  steamrt4_build:
+    runs-on: ${{ matrix.arch }}
+    strategy:
+      matrix:
+        arch: [[self-hosted, ARM64, distrobox]]
+      fail-fast: false
+
+    steps:
+    - uses: actions/checkout@v3
+
+    - name: Set runner label
+      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+
+    - name : submodule checkout
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
+
+    - name: Clean Build Environment
+      run: |
+        rm -Rf ${{runner.workspace}}/build
+        cmake -E make_directory ${{runner.workspace}}/build
+
+    # Setup everything required.
+    - name : distrobox setup
+      run: |
+        distrobox create -Y -i registry.gitlab.steamos.cloud/steamrt/steamrt4/sdk/arm64:4.0.20251117.183306 steamrt4 || true
+        distrobox upgrade steamrt4
+        distrobox enter --name steamrt4 -- sudo apt-get install -y \
+          git cmake ninja-build ccache \
+          lld clang \
+          libclang-dev llvm-dev \
+          libstdc++-14-dev-i386-cross libgcc-14-dev-i386-cross \
+          libstdc++-14-dev-amd64-cross libgcc-14-dev-amd64-cross
+
+    - name: Create Build Environment
+      run: distrobox enter --name steamrt4 -- cmake -E make_directory ${{runner.workspace}}/build
+
+    - name: Configure CMake
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: distrobox enter --name steamrt4 -- cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DBUILD_STEAM_SUPPORT=True -DENABLE_LTO=True -DENABLE_ASSERTIONS=False -DBUILD_THUNKS=True -DBUILD_FEXCONFIG=False -DBUILD_TESTING=False -DENABLE_CLANG_THUNKS=True -DUSE_LINKER=lld -DCMAKE_INSTALL_PREFIX=/usr
+
+    - name: Build
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: distrobox enter --name steamrt4 -- cmake --build . --config $BUILD_TYPE
+
+    - name: install
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      env:
+        DESTDIR: ${{runner.workspace}}/install
+      run: distrobox enter --name steamrt4 -- cmake --build . --config $BUILD_TYPE -t install
+    - name: Upload libraries
+      uses: 'actions/upload-artifact@v4'
+      timeout-minutes: 1
+      with:
+        overwrite: true
+        name: steamrt4_steampipe_depot
+        path: ${{runner.workspace}}/install/*
+        retention-days: 1
+        compression-level: 9

CMakeLists.txt

@@ -33,6 +33,7 @@ option(ENABLE_FEXCORE_PROFILER "Enables use of the FEXCore timeline profiling ca
 set (FEXCORE_PROFILER_BACKEND "gpuvis" CACHE STRING "Set which backend to use for the FEXCore profiler (gpuvis, tracy)")
 option(ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT "Enables glibc memory allocation hooking with fault for CI testing")
 option(USE_PDB_DEBUGINFO "Builds debug info in PDB format" FALSE)
+option(BUILD_STEAM_SUPPORT "Builds FEX for integration into Steam" FALSE)
 
 set (X86_32_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/toolchain_x86_32.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting i686")
 set (X86_64_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/toolchain_x86_64.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting x86_64")
@@ -64,6 +65,10 @@ if (NOT MINGW_BUILD)
   endif()
 endif()
 
+if (BUILD_STEAM_SUPPORT)
+  add_definitions(-DFEX_STEAM_SUPPORT=1)
+endif()
+
 if (ENABLE_FEXCORE_PROFILER)
   add_definitions(-DENABLE_FEXCORE_PROFILER=1)
   string(TOUPPER "${FEXCORE_PROFILER_BACKEND}" FEXCORE_PROFILER_BACKEND)
@@ -319,7 +324,7 @@ if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
 endif()
 
 find_package(PkgConfig REQUIRED)
-find_package(Python 3.0 REQUIRED COMPONENTS Interpreter)
+find_package(Python 3.9 REQUIRED COMPONENTS Interpreter)
 
 set(BUILD_SHARED_LIBS OFF)
 
@@ -476,13 +481,16 @@ add_subdirectory(FEXHeaderUtils/)
 add_subdirectory(CodeEmitter/)
 add_subdirectory(FEXCore/)
 
-if (_M_ARM_64 AND NOT MINGW_BUILD)
+if (_M_ARM_64 AND NOT MINGW_BUILD AND NOT BUILD_STEAM_SUPPORT)
   # Binfmt_misc files must be installed prior to Source/ installs
   add_subdirectory(Data/binfmts/)
 endif()
 
 add_subdirectory(Source/)
-add_subdirectory(Data/AppConfig/)
+
+if (NOT BUILD_STEAM_SUPPORT)
+  add_subdirectory(Data/AppConfig/)
+endif()
 
 # Install the ThunksDB file
 file(GLOB CONFIG_SOURCES CONFIGURE_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/Data/*.json)
@@ -582,6 +590,10 @@ if (BUILD_THUNKS)
   add_dependencies(uninstall uninstall_guest-libs-32)
 endif()
 
+if (BUILD_STEAM_SUPPORT)
+  add_subdirectory(Source/Steam/)
+endif()
+
 set(FEX_VERSION_MAJOR "0")
 set(FEX_VERSION_MINOR "0")
 set(FEX_VERSION_PATCH "0")

CodeEmitter/CodeEmitter/ALUOps.inl

@@ -38,9 +38,7 @@ public:
 
   [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(IsADRRange(Imm), "Unscaled offset too large");
-
-    if (IsADRRange(Imm)) [[likely]] {
+    if (IsADRRange(Imm)) {
       constexpr uint32_t Op = 0b0001'0000 << 24;
       DataProcessing_PCRel_Imm(Op, rd, Imm);
       return BranchEncodeSucceeded::Success;
@@ -73,9 +71,8 @@ public:
 
   [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, const BackwardLabel* Label) {
     int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
-    LOGMAN_THROW_A_FMT(IsADRPRange(Imm) && IsADRPAligned(Imm), "Unscaled offset too large");
 
-    if (IsADRPRange(Imm) && IsADRPAligned(Imm)) [[likely]] {
+    if (IsADRPRange(Imm) && IsADRPAligned(Imm)) {
       constexpr uint32_t Op = 0b1001'0000 << 24;
       DataProcessing_PCRel_Imm(Op, rd, Imm);
       return BranchEncodeSucceeded::Success;
@@ -103,16 +100,22 @@ public:
   }
 
   [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, const BackwardLabel* Label) {
-    int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>());
+    const auto SLocation = reinterpret_cast<int64_t>(Label->Location);
+    const auto ULocation = std::bit_cast<uint64_t>(SLocation);
+
+    const int64_t Imm = SLocation - (GetCursorAddress<int64_t>());
+    const auto UImm = std::bit_cast<uint64_t>(Imm);
+
     if (IsADRRange(Imm)) {
       // If the range is in ADR range then we can just use ADR.
       return adr(rd, Label);
-    } else if (IsADRPRange(Imm)) {
-      int64_t ADRPImm = (reinterpret_cast<int64_t>(Label->Location) & ~0xFFFLL) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
+    }
+    if (IsADRPRange(Imm)) {
+      const int64_t ADRPImm = (SLocation & ~0xFFFLL) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
 
       // If the range is in the ADRP range then we can use ADRP.
-      bool NeedsOffset = !IsADRPAligned(reinterpret_cast<uint64_t>(Label->Location));
-      uint64_t AlignedOffset = reinterpret_cast<uint64_t>(Label->Location) & 0xFFFULL;
+      const bool NeedsOffset = !IsADRPAligned(ULocation);
+      const uint64_t AlignedOffset = ULocation & 0xFFFULL;
 
       // First emit ADRP
       adrp(rd, ADRPImm >> 12);
@@ -125,14 +128,19 @@ public:
       return BranchEncodeSucceeded::Success;
     }
 
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    // Stinky path, we need to load the address as a sequence of movz+movk+movk
+    movz(ARMEmitter::Size::i64Bit, rd, (UImm >> 32) & 0xFFFF, 32);
+    movk(ARMEmitter::Size::i64Bit, rd, (UImm >> 16) & 0xFFFF, 16);
+    movk(ARMEmitter::Size::i64Bit, rd, UImm & 0xFFFF);
+
+    return BranchEncodeSucceeded::Success;
   }
   [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::LONG_ADDRESS_GEN});
-    // Emit a register index and a nop. These will be backpatched.
+    // Emit a register index and two nops. These will be backpatched.
     dc32(rd.Idx());
     nop();
+    nop();
 
     // Forward label doesn't know if it can encode until Bind.
     return BranchEncodeSucceeded::Success;

CodeEmitter/CodeEmitter/BranchOps.inl

@@ -22,7 +22,7 @@ public:
   }
   [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
       constexpr uint32_t Op = 0b0101'010 << 25;
       Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
       return BranchEncodeSucceeded::Success;
@@ -55,7 +55,7 @@ public:
   }
   [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
       constexpr uint32_t Op = 0b0101'010 << 25;
       Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
       return BranchEncodeSucceeded::Success;
@@ -116,7 +116,7 @@ public:
   }
   [[nodiscard]] BranchEncodeSucceeded b(const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) {
       constexpr uint32_t Op = 0b0001'01 << 26;
       UnconditionalBranch(Op, Imm >> 2);
       return BranchEncodeSucceeded::Success;
@@ -151,7 +151,7 @@ public:
 
   [[nodiscard]] BranchEncodeSucceeded bl(const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) {
       constexpr uint32_t Op = 0b1001'01 << 26;
       UnconditionalBranch(Op, Imm >> 2);
 
@@ -189,7 +189,7 @@ public:
   [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
 
-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
       constexpr uint32_t Op = 0b0011'0100 << 24;
       CompareAndBranch(Op, s, rt, Imm >> 2);
       return BranchEncodeSucceeded::Success;
@@ -227,7 +227,7 @@ public:
   [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
 
-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
       constexpr uint32_t Op = 0b0011'0101 << 24;
       CompareAndBranch(Op, s, rt, Imm >> 2);
       return BranchEncodeSucceeded::Success;
@@ -265,7 +265,7 @@ public:
   [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
 
-    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) {
       constexpr uint32_t Op = 0b0011'0110 << 24;
       TestAndBranch(Op, rt, Bit, Imm >> 2);
       return BranchEncodeSucceeded::Success;
@@ -301,9 +301,8 @@ public:
   }
   [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
 
-    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) {
       constexpr uint32_t Op = 0b0011'0111 << 24;
       TestAndBranch(Op, rt, Bit, Imm >> 2);
       return BranchEncodeSucceeded::Success;

CodeEmitter/CodeEmitter/Emitter.h

@@ -586,6 +586,10 @@ concept IsXOrWRegister = std::is_same_v<T, XRegister> || std::is_same_v<T, WRegi
 template<typename T>
 concept IsQOrDRegister = std::is_same_v<T, QRegister> || std::is_same_v<T, DRegister>;
 
+template<typename T>
+concept IsLabel = std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
+                  std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>;
+
 enum class BranchEncodeSucceeded {
   Success,
   Failure,
@@ -658,7 +662,7 @@ public:
     case ForwardLabel::InstType::ADR: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      if (!IsADRRange(Imm)) [[unlikely]] {
+      if (!IsADRRange(Imm)) {
         // Can't bind.
         return false;
       }
@@ -674,7 +678,7 @@ public:
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
 
-      if (!(IsADRPRange(Imm) && IsADRPAligned(Imm))) [[unlikely]] {
+      if (!(IsADRPRange(Imm) && IsADRPAligned(Imm))) {
         // Can't bind.
         return false;
       }
@@ -691,7 +695,7 @@ public:
     case ForwardLabel::InstType::B: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      if (!(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0))) [[unlikely]] {
+      if (!(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0))) {
         // Can't bind.
         return false;
       }
@@ -707,7 +711,7 @@ public:
     case ForwardLabel::InstType::TEST_BRANCH: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      if (!(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0))) [[unlikely]] {
+      if (!(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0))) {
         // Can't bind.
         return false;
       }
@@ -724,7 +728,7 @@ public:
     case ForwardLabel::InstType::RELATIVE_LOAD: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      if (!(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0))) [[unlikely]] {
+      if (!(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0))) {
         // Can't bind.
         return false;
       }
@@ -737,38 +741,44 @@ public:
       break;
     }
     case ForwardLabel::InstType::LONG_ADDRESS_GEN: {
-      uint32_t* Instructions = reinterpret_cast<uint32_t*>(Label->Location);
-      int64_t ImmInstOne = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[0]);
-      int64_t ImmInstTwo = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[1]);
-      auto OriginalOffset = GetCursorOffset();
+      const auto* Instructions = reinterpret_cast<uint32_t*>(Label->Location);
+      const auto ImmInstOne = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[0]);
+      const auto ImmInstTwo = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[1]);
+      const auto ImmInstThree = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[2]);
+      const auto OriginalOffset = GetCursorOffset();
 
-      auto InstOffset = GetCursorOffsetFromAddress(Instructions);
+      const auto InstOffset = GetCursorOffsetFromAddress(Instructions);
       SetCursorOffset(InstOffset);
 
       // We encoded the destination register in to the first instruction space.
       // Read it back.
       ARMEmitter::Register DestReg(Instructions[0]);
 
-      if (IsADRRange(ImmInstTwo)) {
-        // If within ADR range from the second instruction, then we can emit NOP+ADR
+      if (IsADRRange(ImmInstThree)) {
+        // If within ADR range from the third instruction, then we can emit NOP+NOP+ADR
         nop();
-        adr(DestReg, static_cast<uint32_t>(ImmInstTwo) & 0x7FFF);
-      } else if (IsADRPRange(ImmInstOne)) {
+        nop();
+        adr(DestReg, static_cast<uint32_t>(ImmInstThree) & 0x7FFF);
+      } else if (IsADRPRange(ImmInstTwo)) {
 
         // If within ADRP range from the first instruction, then we are /definitely/ in range for the second instruction.
         // First check if we are in non-offset range for second instruction.
         if (IsADRPAligned(reinterpret_cast<uint64_t>(CurrentAddress))) {
-          // We can emit nop + adrp
+          // We can emit nop + nop + adrp
+          nop();
+          nop();
+          adrp(DestReg, static_cast<uint32_t>(ImmInstThree >> 12) & 0x7FFF);
+        } else {
+          // Not aligned, need nop + adrp + add
           nop();
           adrp(DestReg, static_cast<uint32_t>(ImmInstTwo >> 12) & 0x7FFF);
-        } else {
-          // Not aligned, need adrp + add
-          adrp(DestReg, static_cast<uint32_t>(ImmInstOne >> 12) & 0x7FFF);
-          add(ARMEmitter::Size::i64Bit, DestReg, DestReg, ImmInstOne & 0xFFF);
+          add(ARMEmitter::Size::i64Bit, DestReg, DestReg, ImmInstTwo & 0xFFF);
         }
       } else {
-        LOGMAN_MSG_A_FMT("Unscaled offset is too large");
-        FEX_UNREACHABLE;
+        // Stinky path, we need to emit a movz+movk+movk sequence.
+        movz(ARMEmitter::Size::i64Bit, DestReg, uint32_t(ImmInstOne >> 32) & 0x7FFF, 32);
+        movk(ARMEmitter::Size::i64Bit, DestReg, uint32_t(ImmInstOne >> 16) & 0xFFFF, 16);
+        movk(ARMEmitter::Size::i64Bit, DestReg, uint32_t(ImmInstOne) & 0xFFFF);
       }
 
       SetCursorOffset(OriginalOffset);

CodeEmitter/CodeEmitter/SVEOps.inl

@@ -5125,7 +5125,7 @@ private:
   requires (std::is_same_v<T, float> || std::is_same_v<T, double>)
   [[nodiscard]]
   static bool IsValidFPValueForImm8(T value) {
-    const uint64_t bits = FEXCore::BitCast<FloatToEquivalentUInt<T>>(value);
+    const uint64_t bits = std::bit_cast<FloatToEquivalentUInt<T>>(value);
     const uint64_t datasize_idx = FEXCore::ilog2(sizeof(T)) - 1;
 
     static constexpr std::array mantissa_masks {
@@ -5171,7 +5171,7 @@ protected:
     LOGMAN_THROW_A_FMT(IsValidFPValueForImm8(value), "Value ({}) cannot be encoded into an 8-bit immediate", value);
 #endif
 
-    const auto bits = FEXCore::BitCast<uint32_t>(value);
+    const auto bits = std::bit_cast<uint32_t>(value);
     const auto sign = (bits & 0x80000000) >> 24;
     const auto expb2 = (bits & 0x20000000) >> 23;
     const auto b5_to_0 = (bits >> 19) & 0x3F;
@@ -5184,7 +5184,7 @@ protected:
     LOGMAN_THROW_A_FMT(IsValidFPValueForImm8(value), "Value ({}) cannot be encoded into an 8-bit immediate", value);
 #endif
 
-    const auto bits = FEXCore::BitCast<uint64_t>(value);
+    const auto bits = std::bit_cast<uint64_t>(value);
     const auto sign = (bits & 0x80000000'00000000) >> 56;
     const auto expb2 = (bits & 0x20000000'00000000) >> 55;
     const auto b5_to_0 = (bits >> 48) & 0x3F;

Data/nix/WineOnArm/shell.nix

@@ -2,8 +2,8 @@
 
 let
   toolchain = pkgs.fetchzip {
-    url = "https://github.com/bylaws/llvm-mingw/releases/download/20250305/llvm-mingw-20250305-ucrt-ubuntu-20.04-aarch64.tar.xz";
-    sha256 = "sha256-cA03/ab9O61eO9+S2JzIXD4V0HzTXK5/AYyxW2d73Po=";
+    url = "https://github.com/bylaws/llvm-mingw/releases/download/20250920/llvm-mingw-20250920-ucrt-ubuntu-22.04-aarch64.tar.xz";
+    sha256 = "sha256-LaojKjC8KzY+soW5u6eoDoXE3qtYk9Ejr7M3enTqRAE=";
   };
 
   cmakeToolchainFile = pkgs.substitute {

FEXCore/CMakeLists.txt

@@ -74,9 +74,11 @@ add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-fno-strict-aliasing> $<$<COMPILE_
 
 add_subdirectory(Source/)
 
-install (DIRECTORY include/FEXCore ${CMAKE_BINARY_DIR}/include/FEXCore
-  DESTINATION include
-  COMPONENT Development)
+if (NOT BUILD_STEAM_SUPPORT)
+  install (DIRECTORY include/FEXCore ${CMAKE_BINARY_DIR}/include/FEXCore
+    DESTINATION include
+    COMPONENT Development)
+endif()
 
 if (BUILD_TESTING)
   add_subdirectory(unittests/)

FEXCore/Scripts/config_generator.py

@@ -200,6 +200,15 @@ def print_man_environment_tail():
     ],
     "''", True)
 
+    print_man_env_option(
+    "APP_CACHE_LOCATION",
+    [
+    "Allows the user to override where FEX stores and loads cache files",
+    "By default FEX will look in $XDG_CACHE_HOME/fex-emu/ or $HOME/.cache/fex-emu/",
+    "This will override the full path, trailing forward-slash is expected to exist",
+    ],
+    "''", True)
+
 def print_man_header():
     header ='''.Dd {0}
 .Dt FEX

FEXCore/Scripts/json_ir_generator.py

@@ -58,10 +58,10 @@ class OpDefinition:
     JITDispatch: bool
     JITDispatchOverride: str
     TiedSource: int
-    Inline: list
-    Arguments: list
-    EmitValidation: list
-    Desc: list
+    Inline: list[str]
+    Arguments: list[OpArgument]
+    EmitValidation: list[str]
+    Desc: list[str]
 
     def __init__(self):
         self.Name = None
@@ -92,19 +92,14 @@ class OpDefinition:
         attrs = vars(self)
         print(", ".join("%s: %s" % item for item in attrs.items()))
 
-IRTypesToCXX = {}
-CXXTypeToIR = {}
-IROps = []
+IRTypesToCXX: dict[str, IRType] = {}
+CXXTypeToIR: dict[str, IRType] = {}
+IROps: list[OpDefinition] = []
 
-IROpNameMap = {}
+IROpNameSet: set[str] = set()
 
-def is_ssa_type(type):
-    if (type == "SSA" or
-       type == "GPR" or
-       type == "GPRPair" or
-       type == "FPR"):
-       return True
-    return False
+def is_ssa_type(op_type: str):
+    return op_type in {"SSA", "GPR", "GPRPair", "FPR"}
 
 def parse_irtypes(irtypes):
     for op_key, op_val in irtypes.items():
@@ -219,11 +214,8 @@ def parse_ops(ops):
                         OpArg.DefaultInitializer = DefaultInit[1][:-1]
 
                     # If SSA type then we can generate validation for this op
-                    if (OpArg.IsSSA and
-                        (OpArg.Type == "GPR" or
-                        OpArg.Type == "GPRPair" or
-                        OpArg.Type == "FPR")):
-                        OpDef.EmitValidation.append(f"GetOpRegClass({ArgName}) == InvalidClass || WalkFindRegClass({ArgName}) == {OpArg.Type}Class")
+                    if OpArg.IsSSA and OpArg.Type in {"GPR", "GPRPair", "FPR"}:
+                        OpDef.EmitValidation.append(f"GetOpRegClass({ArgName}) == RegClass::Invalid || WalkFindRegClass({ArgName}) == RegClass::{OpArg.Type}")
 
                     OpArg.Name = ArgName
                     OpArg.NameWithPrefix = NameWithPrefix
@@ -296,22 +288,29 @@ def parse_ops(ops):
             #OpDef.print()
 
             # Error on duplicate op
-            if OpDef.Name in IROpNameMap:
+            if OpDef.Name in IROpNameSet:
                 ExitError("Duplicate Op defined! {}".format(OpDef.Name))
 
             IROps.append(OpDef)
-            IROpNameMap[OpDef.Name] = 1
+            IROpNameSet.add(OpDef.Name)
 
 # Print out enum values
-def print_enums():
+def print_enums(enums):
     output_file.write("#ifdef IROP_ENUM\n")
     output_file.write("enum IROps : uint16_t {\n")
-
     for op in IROps:
         output_file.write("\tOP_{},\n" .format(op.Name.upper()))
-
     output_file.write("};\n")
 
+    for name, members in enums.items():
+        output_file.write(f"enum {name} {{\n")
+        for member in members:
+            if member:
+                output_file.write(f"\t{member}\n")
+            else:
+                output_file.write("\n")
+        output_file.write("};\n\n")
+
     output_file.write("#undef IROP_ENUM\n")
     output_file.write("#endif\n\n")
 
@@ -408,7 +407,7 @@ def print_ir_sizes():
     [[nodiscard, gnu::const]] std::string_view const& GetName(IROps Op);
     [[nodiscard, gnu::const]] uint8_t GetArgs(IROps Op);
     [[nodiscard, gnu::const]] uint8_t GetRAArgs(IROps Op);
-    [[nodiscard, gnu::const]] FEXCore::IR::RegisterClassType GetRegClass(IROps Op);
+    [[nodiscard, gnu::const]] FEXCore::IR::RegClass GetRegClass(IROps Op);
     [[nodiscard, gnu::const]] bool HasSideEffects(IROps Op);
     [[nodiscard, gnu::const]] bool ImplicitFlagClobber(IROps Op);
     [[nodiscard, gnu::const]] bool GetHasDest(IROps Op);
@@ -422,30 +421,29 @@ def print_ir_sizes():
 def print_ir_reg_classes():
     output_file.write("#ifdef IROP_REG_CLASSES_IMPL\n")
 
-    output_file.write("constexpr std::array<FEXCore::IR::RegisterClassType, IROps::OP_LAST + 1> IRRegClasses = {\n")
+    output_file.write("constexpr std::array<FEXCore::IR::RegClass, IROps::OP_LAST + 1> IRRegClasses = {\n")
     for op in IROps:
         if op.Name == "Last":
-            output_file.write("\tFEXCore::IR::InvalidClass,\n")
+            output_file.write("\tRegClass::Invalid,\n")
         else:
-            Class = "Invalid"
-            if op.HasDest and op.DestType == None:
+            if op.HasDest and op.DestType is None:
                 ExitError("IR op {} has destination with no destination class".format(op.Name))
 
             if op.HasDest and op.DestType == "SSA": # Special case SSA type
-                output_file.write("\tFEXCore::IR::ComplexClass,\n")
+                output_file.write("\tRegClass::Complex,\n")
             elif op.HasDest:
-                output_file.write("\tFEXCore::IR::{}Class,\n".format(op.DestType))
+                output_file.write("\tRegClass::{},\n".format(op.DestType))
             else:
                 # No destination so it has an invalid destination class
-                output_file.write("\tFEXCore::IR::InvalidClass, // No destination\n")
+                output_file.write("\tRegClass::Invalid, // No destination\n")
 
 
     output_file.write("};\n\n")
 
     output_file.write("// Make sure our array maps directly to the IROps enum\n")
-    output_file.write("static_assert(IRRegClasses[IROps::OP_LAST] == FEXCore::IR::InvalidClass);\n\n")
+    output_file.write("static_assert(IRRegClasses[IROps::OP_LAST] == RegClass::Invalid);\n\n")
 
-    output_file.write("FEXCore::IR::RegisterClassType GetRegClass(IROps Op) { return IRRegClasses[Op]; }\n\n")
+    output_file.write("FEXCore::IR::RegClass GetRegClass(IROps Op) { return IRRegClasses[Op]; }\n\n")
 
     output_file.write("#undef IROP_REG_CLASSES_IMPL\n")
     output_file.write("#endif\n\n")
@@ -568,9 +566,7 @@ def print_ir_arg_printer():
 
             SSAArgNum = 0
             FirstArg = True
-            for i in range(0, len(op.Arguments)):
-                arg = op.Arguments[i]
-
+            for arg in op.Arguments:
                 # No point printing temporaries that we can't recover
                 if arg.Temporary:
                     continue
@@ -671,7 +667,7 @@ def print_ir_allocator_helpers():
     output_file.write("\t\treturn HeaderOp->Op;\n")
     output_file.write("\t}\n\n")
 
-    output_file.write("\tFEXCore::IR::RegisterClassType GetOpRegClass(const OrderedNode *Op) const {\n")
+    output_file.write("\tFEXCore::IR::RegClass GetOpRegClass(const OrderedNode *Op) const {\n")
     output_file.write("\t\treturn GetRegClass(GetOpType(Op));\n")
     output_file.write("\t}\n\n")
 
@@ -685,22 +681,21 @@ def print_ir_allocator_helpers():
             output_file.write("\tIRPair<IROp_{}> _{}(" .format(op.Name, op.Name))
 
             # Output SSA args first
-            for i in range(0, len(op.Arguments)):
-                arg = op.Arguments[i]
-                LastArg = len(op.Arguments) - i - 1 == 0
+            for i, arg in enumerate(op.Arguments):
+                LastArg = i == len(op.Arguments) - 1
 
                 if arg.Temporary:
                     CType = IRTypesToCXX[arg.Type].CXXName
-                    output_file.write("{} {}".format(CType, arg.Name));
+                    output_file.write("{} {}".format(CType, arg.Name))
                 elif arg.IsSSA:
                     # SSA value
                     output_file.write("OrderedNodeWrapper {}".format(arg.Name))
                 else:
                     # User defined op that is stored
                     CType = IRTypesToCXX[arg.Type].CXXName
-                    output_file.write("{} {}".format(CType, arg.Name));
+                    output_file.write("{} {}".format(CType, arg.Name))
 
-                if arg.DefaultInitializer != None:
+                if arg.DefaultInitializer:
                     output_file.write(" = {}".format(arg.DefaultInitializer))
 
                 if not LastArg:
@@ -758,20 +753,19 @@ def print_ir_allocator_helpers():
             if op.SSAArgNum:
                 output_file.write("\tIRPair<IROp_{}> _{}(" .format(op.Name, op.Name))
 
-                for i in range(0, len(op.Arguments)):
-                    arg = op.Arguments[i]
-                    LastArg = len(op.Arguments) - i - 1 == 0
+                for i, arg in enumerate(op.Arguments):
+                    LastArg = i == len(op.Arguments) - 1
 
                     if arg.Temporary:
                         CType = IRTypesToCXX[arg.Type].CXXName
-                        output_file.write("{} {}".format(CType, arg.Name));
+                        output_file.write("{} {}".format(CType, arg.Name))
                     elif arg.IsSSA:
                         output_file.write("OrderedNode *{}".format(arg.Name))
                     else:
                         CType = IRTypesToCXX[arg.Type].CXXName
-                        output_file.write("{} {}".format(CType, arg.Name));
+                        output_file.write("{} {}".format(CType, arg.Name))
 
-                    if arg.DefaultInitializer != None:
+                    if arg.DefaultInitializer:
                         output_file.write(" = {}".format(arg.DefaultInitializer))
 
                     if not LastArg:
@@ -812,16 +806,15 @@ def print_ir_allocator_helpers():
                 print_validation(op)
 
                 output_file.write(f"\t\treturn _{op.Name}(")
-                for i in range(0, len(op.Arguments)):
-                    arg = op.Arguments[i]
-                    LastArg = len(op.Arguments) - i - 1 == 0
+                for i, arg in enumerate(op.Arguments):
+                    LastArg = i == len(op.Arguments) - 1
                     output_file.write(arg.Name)
                     if arg.IsSSA:
                         output_file.write("->Wrapped(ListDataBegin)")
                     if not LastArg:
                         output_file.write(", ")
-                output_file.write(");\n");
-                output_file.write("\t}\n\n");
+                output_file.write(");\n")
+                output_file.write("\t}\n\n")
 
     output_file.write("#undef IROP_ALLOCATE_HELPERS\n")
     output_file.write("#endif\n")
@@ -852,8 +845,8 @@ def print_ir_dispatcher_dispatch():
     output_dispatch_file.write("#endif\n")
 
 
-if (len(sys.argv) < 4):
-    ExitError()
+if len(sys.argv) < 4:
+    ExitError("Insufficient parameters passed to script")
 
 output_filename = sys.argv[2]
 output_dispatcher_filename = sys.argv[3]
@@ -865,6 +858,7 @@ json_file.close()
 json_object = json.loads(json_text)
 json_object = {k.upper(): v for k, v in json_object.items()}
 
+enums = json_object["ENUMS"]
 ops = json_object["OPS"]
 irtypes = json_object["IRTYPES"]
 defines = json_object["DEFINES"]
@@ -874,7 +868,7 @@ parse_ops(ops)
 
 output_file = open(output_filename, "w")
 
-print_enums()
+print_enums(enums)
 print_ir_structs(defines)
 print_ir_sizes()
 print_ir_reg_classes()

FEXCore/Source/CMakeLists.txt

@@ -31,7 +31,6 @@ set (SRCS
   Interface/Core/OpcodeDispatcher/X87.cpp
   Interface/Core/OpcodeDispatcher/X87F64.cpp
   Interface/Core/OpcodeDispatcher.cpp
-  Interface/Core/X86HelperGen.cpp
   Interface/Core/ArchHelpers/Arm64Emitter.cpp
   Interface/Core/Dispatcher/Dispatcher.cpp
   Interface/Core/Interpreter/Fallbacks/InterpreterFallbacks.cpp
@@ -202,8 +201,10 @@ add_custom_target(CONFIG_INC
   DEPENDS "${OUTPUT_MAN_NAME}"
   DEPENDS "${OUTPUT_MAN_NAME_COMPRESS}")
 
-# Install the compressed man page
-install(FILES ${OUTPUT_MAN_NAME_COMPRESS} COMPONENT Runtime DESTINATION ${MAN_DIR}/man1)
+if (NOT BUILD_STEAM_SUPPORT)
+  # Install the compressed man page
+  install(FILES ${OUTPUT_MAN_NAME_COMPRESS} COMPONENT Runtime DESTINATION ${MAN_DIR}/man1)
+endif()
 
 # Add in diagnostic colours if the option is available.
 # Ninja code generator will kill colours if this isn't here
@@ -290,7 +291,7 @@ AddObject(${PROJECT_NAME}_object OBJECT)
 AddLibrary(${PROJECT_NAME} STATIC)
 AddLibrary(${PROJECT_NAME}_shared SHARED)
 
-if (NOT MINGW_BUILD)
+if (NOT MINGW_BUILD AND NOT BUILD_STEAM_SUPPORT)
   install(TARGETS ${PROJECT_NAME}_shared
     LIBRARY
       DESTINATION ${CMAKE_INSTALL_LIBDIR}

FEXCore/Source/Common/JitSymbols.h

@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: MIT
 #pragma once
+#include <FEXCore/Utils/TypeDefines.h>
 
 #include <FEXCore/fextl/memory.h>
 
@@ -12,7 +13,7 @@ namespace FEXCore {
 // Buffered JIT symbol tracking.
 struct JITSymbolBuffer {
   // Maximum buffer size to ensure we are a page in size.
-  constexpr static size_t BUFFER_SIZE = 4096 - (8 * 2);
+  constexpr static size_t BUFFER_SIZE = FEXCore::Utils::FEX_PAGE_SIZE - (8 * 2);
   // Maximum distance until the end of the buffer to do a write.
   constexpr static size_t NEEDS_WRITE_DISTANCE = BUFFER_SIZE - 64;
   // Maximum time threshhold to wait before a buffer write occurs.
@@ -27,7 +28,7 @@ struct JITSymbolBuffer {
   size_t Offset {};
   char Buffer[BUFFER_SIZE] {};
 };
-static_assert(sizeof(JITSymbolBuffer) == 4096, "Ensure this is one page in size");
+static_assert(sizeof(JITSymbolBuffer) == FEXCore::Utils::FEX_PAGE_SIZE, "Ensure this is one page in size");
 
 class JITSymbols final {
 public:

FEXCore/Source/Common/SoftFloat.h

@@ -4,9 +4,9 @@
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/fextl/sstream.h>
 #include <FEXCore/fextl/string.h>
-#include <FEXHeaderUtils/BitUtils.h>
 #include "cephes_128bit.h"
 
+#include <bit>
 #include <cmath>
 #include <cstring>
 #include <stdint.h>
@@ -501,12 +501,12 @@ struct FEX_PACKED X80SoftFloat {
 
   float ToF32(softfloat_state* state) const {
     const float32_t Result = extF80_to_f32(state, *this);
-    return FEXCore::BitCast<float>(Result);
+    return std::bit_cast<float>(Result);
   }
 
   double ToF64(softfloat_state* state) const {
     const float64_t Result = extF80_to_f64(state, *this);
-    return FEXCore::BitCast<double>(Result);
+    return std::bit_cast<double>(Result);
   }
 
   FEXCore::VectorRegType ToVector() const {
@@ -518,7 +518,7 @@ struct FEX_PACKED X80SoftFloat {
   BIGFLOAT ToFMax(softfloat_state* state) const {
 #if BIGFLOATSIZE == 16
     const float128_t Result = extF80_to_f128(state, *this);
-    return FEXCore::BitCast<BIGFLOAT>(Result);
+    return std::bit_cast<BIGFLOAT>(Result);
 #else
     BIGFLOAT result {};
     memcpy(&result, this, sizeof(result));
@@ -577,18 +577,18 @@ struct FEX_PACKED X80SoftFloat {
   }
 
   X80SoftFloat(softfloat_state* state, const float rhs) {
-    *this = f32_to_extF80(state, FEXCore::BitCast<float32_t>(rhs));
+    *this = f32_to_extF80(state, std::bit_cast<float32_t>(rhs));
   }
 
   X80SoftFloat(softfloat_state* state, const double rhs) {
-    *this = f64_to_extF80(state, FEXCore::BitCast<float64_t>(rhs));
+    *this = f64_to_extF80(state, std::bit_cast<float64_t>(rhs));
   }
 
   X80SoftFloat(softfloat_state* state, BIGFLOAT rhs) {
 #if BIGFLOATSIZE == 16
-    *this = f128_to_extF80(state, FEXCore::BitCast<float128_t>(rhs));
+    *this = f128_to_extF80(state, std::bit_cast<float128_t>(rhs));
 #else
-    *this = FEXCore::BitCast<long double>(rhs);
+    *this = std::bit_cast<long double>(rhs);
 #endif
   }
 

FEXCore/Source/Common/StringConv.h

@@ -2,59 +2,23 @@
 #pragma once
 #include <FEXCore/fextl/string.h>
 
-#include <cstdint>
+#include <concepts>
 #include <string_view>
-#include <optional>
 
 namespace FEXCore::StrConv {
-inline bool Conv(std::string_view Value, bool* Result) {
-  *Result = std::strtoull(Value.data(), nullptr, 0);
+template<std::integral T>
+bool Conv(std::string_view Value, T* Result) {
+  if constexpr (std::is_signed_v<T>) {
+    *Result = static_cast<T>(std::strtoll(Value.data(), nullptr, 0));
+  } else {
+    *Result = static_cast<T>(std::strtoull(Value.data(), nullptr, 0));
+  }
   return true;
 }
 
-inline bool Conv(std::string_view Value, uint8_t* Result) {
-  *Result = std::strtoul(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, int8_t* Result) {
-  *Result = std::strtol(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, uint16_t* Result) {
-  *Result = std::strtoul(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, int16_t* Result) {
-  *Result = std::strtol(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, uint32_t* Result) {
-  *Result = std::strtoul(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, int32_t* Result) {
-  *Result = std::strtol(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, uint64_t* Result) {
-  *Result = std::strtoull(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, int64_t* Result) {
-  *Result = std::strtoll(Value.data(), nullptr, 0);
-  return true;
-}
-
-template<typename T, typename = std::enable_if<std::is_enum<T>::value, T>>
-inline bool Conv(std::string_view Value, T* Result) {
-  *Result = static_cast<T>(std::stoull(Value.data(), nullptr, 0));
+template<typename T, typename = std::enable_if_t<std::is_enum_v<T>, T>>
+bool Conv(std::string_view Value, T* Result) {
+  *Result = static_cast<T>(std::strtoull(Value.data(), nullptr, 0));
   return true;
 }
 

FEXCore/Source/Interface/Config/Config.cpp

@@ -30,14 +30,14 @@ class Context;
 }
 
 namespace FEXCore::Config {
-namespace DefaultValues {
+namespace detail {
 #define P(x) x
 #define OPT_BASE(type, group, enum, json, default) const P(type) P(enum) = P(default);
 #define OPT_STR(group, enum, json, default) const std::string_view P(enum) = P(default);
 #define OPT_STRARRAY(group, enum, json, default) OPT_STR(group, enum, json, default)
 #define OPT_STRENUM(group, enum, json, default) const uint64_t P(enum) = FEXCore::ToUnderlying(P(default));
 #include <FEXCore/Config/ConfigValues.inl>
-} // namespace DefaultValues
+} // namespace detail
 
 enum Paths {
   PATH_DATA_DIR_LOCAL = 0,
@@ -134,7 +134,7 @@ public:
   void Load();
 
   template<typename T>
-  requires (!std::is_same_v<fextl::string, T> && !std::is_same_v<DefaultValues::Type::StringArrayType, T>)
+  requires (!std::is_same_v<fextl::string, T> && !std::is_same_v<StringArrayType, T>)
   std::optional<T> GetConv(ConfigOption Option) {
     const auto it = OptionMap.find(Option);
     if (it == OptionMap.end()) {
@@ -142,7 +142,7 @@ public:
     }
 
     const auto& Value = it->second;
-    LOGMAN_THROW_A_FMT(!std::holds_alternative<DefaultValues::Type::StringArrayType>(Value), "Tried to get config of invalid type!");
+    LOGMAN_THROW_A_FMT(!std::holds_alternative<StringArrayType>(Value), "Tried to get config of invalid type!");
 
     if (std::holds_alternative<T>(Value)) [[likely]] {
       return std::get<T>(Value);
@@ -165,7 +165,7 @@ public:
 
 private:
   void MergeConfigMap(const LayerOptions& Options);
-  void MergeEnvironmentVariables(const ConfigOption& Option, const DefaultValues::Type::StringArrayType& Value);
+  void MergeEnvironmentVariables(const ConfigOption& Option, const StringArrayType& Value);
 };
 
 void MetaLayer::Load() {
@@ -181,7 +181,7 @@ void MetaLayer::Load() {
 }
 
 
-void MetaLayer::MergeEnvironmentVariables(const ConfigOption& Option, const DefaultValues::Type::StringArrayType& Value) {
+void MetaLayer::MergeEnvironmentVariables(const ConfigOption& Option, const StringArrayType& Value) {
   // Environment variables need a bit of additional work
   // We want to merge the arrays rather than overwrite entirely
   auto MetaEnvironment = OptionMap.find(Option);
@@ -193,7 +193,7 @@ void MetaLayer::MergeEnvironmentVariables(const ConfigOption& Option, const Defa
 
   // If an environment variable exists in both current meta and in the incoming layer then the meta layer value is overwritten
   fextl::unordered_map<fextl::string, fextl::string> LookupMap;
-  const auto AddToMap = [&LookupMap](const DefaultValues::Type::StringArrayType& Value) {
+  const auto AddToMap = [&LookupMap](const StringArrayType& Value) {
     for (const auto& EnvVar : Value) {
       const auto ItEq = EnvVar.find_first_of('=');
       if (ItEq == fextl::string::npos) {
@@ -209,7 +209,7 @@ void MetaLayer::MergeEnvironmentVariables(const ConfigOption& Option, const Defa
     }
   };
 
-  AddToMap(std::get<DefaultValues::Type::StringArrayType>(MetaEnvironment->second));
+  AddToMap(std::get<StringArrayType>(MetaEnvironment->second));
   AddToMap(Value);
 
   // Now with the two layers merged in the map
@@ -225,8 +225,8 @@ void MetaLayer::MergeConfigMap(const LayerOptions& Options) {
   // Insert this layer's options, overlaying previous options that exist here
   for (auto& it : Options) {
     if (it.first == FEXCore::Config::ConfigOption::CONFIG_ENV || it.first == FEXCore::Config::ConfigOption::CONFIG_HOSTENV) {
-      LOGMAN_THROW_A_FMT(std::holds_alternative<DefaultValues::Type::StringArrayType>(it.second), "Tried to get config of invalid type!");
-      MergeEnvironmentVariables(it.first, std::get<DefaultValues::Type::StringArrayType>(it.second));
+      LOGMAN_THROW_A_FMT(std::holds_alternative<StringArrayType>(it.second), "Tried to get config of invalid type!");
+      MergeEnvironmentVariables(it.first, std::get<StringArrayType>(it.second));
     } else {
       OptionMap.insert_or_assign(it.first, it.second);
     }
@@ -423,7 +423,7 @@ bool Exists(ConfigOption Option) {
   return Meta->OptionExists(Option);
 }
 
-std::optional<DefaultValues::Type::StringArrayType*> All(ConfigOption Option) {
+std::optional<StringArrayType*> All(ConfigOption Option) {
   return Meta->All(Option);
 }
 
@@ -436,6 +436,12 @@ std::optional<T> GetConv(ConfigOption Option) {
   return Meta->GetConv<T>(Option);
 }
 
+template std::optional<bool> GetConv(ConfigOption Option);
+template std::optional<uint8_t> GetConv(ConfigOption Option);
+template std::optional<int32_t> GetConv(ConfigOption Option);
+template std::optional<uint32_t> GetConv(ConfigOption Option);
+template std::optional<uint64_t> GetConv(ConfigOption Option);
+
 void Set(ConfigOption Option, std::string_view Data) {
   Meta->Set(Option, Data);
 }
@@ -491,13 +497,12 @@ template Value<uint8_t>::Value(FEXCore::Config::ConfigOption _Option, uint8_t De
 template Value<uint64_t>::Value(FEXCore::Config::ConfigOption _Option, uint64_t Default);
 
 template<typename T>
-void Value<T>::GetListIfExists(FEXCore::Config::ConfigOption Option, DefaultValues::Type::StringArrayType* List) {
+void Value<T>::GetListIfExists(FEXCore::Config::ConfigOption Option, StringArrayType* List) {
   auto Value = FEXCore::Config::All(Option);
   List->clear();
   if (Value) {
     *List = **Value;
   }
 }
-template void Value<DefaultValues::Type::StringArrayType>::GetListIfExists(FEXCore::Config::ConfigOption Option,
-                                                                           DefaultValues::Type::StringArrayType* List);
+template void Value<StringArrayType>::GetListIfExists(FEXCore::Config::ConfigOption Option, StringArrayType* List);
 } // namespace FEXCore::Config

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

@@ -16,6 +16,13 @@
           "Maximum number of instruction to store in a block"
         ]
       },
+      "EnableCodeCachingWIP": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enable the code caching subsystem"
+        ]
+      },
       "HostFeatures": {
         "Type": "strenum",
         "Default": "FEXCore::Config::HostFeatures::OFF",
@@ -59,7 +66,9 @@
           "ENABLEWFXT": "enablewfxt",
           "DISABLEWFXT": "disablewfxt",
           "ENABLE3DNOW": "enable3dnow",
-          "DISABLE3DNOW": "disable3dnow"
+          "DISABLE3DNOW": "disable3dnow",
+          "ENABLESSE4A": "enablesse4a",
+          "DISABLESSE4A": "disablesse4a"
         },
         "Desc": [
           "Allows controlling of the CPU features in the JIT.",
@@ -82,7 +91,8 @@
           "\t{enable,disable}svebitperm: Will force enable or disable svebitperm even if the host doesn't support it",
           "\t{enable,disable}preserveallabi: Will force enable or disable preserve_all abi even if the host doesn't support it",
           "\t{enable,disable}wfxt: Will force enable or disable wfxt even if the host doesn't support it",
-          "\t{enable,disable}3dnow: Will force enable or disable 3DNow even if the host doesn't support it"
+          "\t{enable,disable}3dnow: Will force enable or disable 3DNow! even if the host doesn't support it",
+          "\t{enable,disable}sse4a: Will force enable or disable SSE4a even if the host doesn't support it"
         ]
       },
       "SmallTSCScale": {
@@ -91,6 +101,13 @@
         "Desc": [
           "Scales the cycle counter on systems that have low frequencies."
         ]
+      },
+      "CPUFeatureRegisters": {
+        "Type": "str",
+        "Default": "",
+        "Desc": [
+          "Allows overriding cpu feature flags for manual testing"
+        ]
       }
     },
     "Emulation": {
@@ -158,6 +175,44 @@
         "Desc": [
           "Allows the user to pass additional arguments to the application"
         ]
+      },
+      "DisableL2Cache": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Disables FEXCore's JIT L2 cache lookup. Saving memory.",
+          "Can potentially introduce more stutters."
+        ]
+      },
+      "DynamicL1Cache": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Switches FEXCore's JIT L1 cache to be dynamically sized. Saving memory.",
+          "Can potentially introduce more stutters."
+        ]
+      },
+      "DynamicL1CacheIncreaseCountHeuristic": {
+        "Type": "uint64",
+        "Default": "250",
+        "Desc": [
+          "Threshold of lookups per second that the L1 dynamic cache should increase its size.",
+          "Lower numbers means more aggressive scaling upward to the maximum size.",
+          "Higher numbers means more conservative scaling, using less memory.",
+          "Can potentially introduce stutters, more likely the higher the number.",
+          "Don't have this number smaller than the decrease count!"
+        ]
+      },
+      "DynamicL1CacheDecreaseCountHeuristic": {
+        "Type": "uint64",
+        "Default": "50",
+        "Desc": [
+          "Threshold of lookups per second that the L1 dynamic cache should decrease its size.",
+          "The higher the number, the more aggressively it reduces the L1 cache size.",
+          "Lower numbers means more conservative memory savings.",
+          "Can potentially introduce more stutters, more likely the higher the number.",
+          "Don't have this number larger than the increase count!"
+        ]
       }
     },
     "Debug": {
@@ -328,11 +383,11 @@
           "Requires a supported version of Mangohud to see the results"
         ]
       },
-      "TraceProfiler": {
+      "EnableGpuvisProfiling": {
         "Type": "bool",
         "Default": "false",
         "Desc": [
-          "Enables FEX's trace profiler. Using gpuvis or tracy"
+          "Enables profiling when FEX was built with the gpuvis profiler backend."
         ]
       }
     },
@@ -388,12 +443,19 @@
           "This is required to ensure a split-lock doesn't tear inside the process"
         ]
       },
+      "KernelUnalignedAtomicBackpatching": {
+        "Type": "bool",
+        "Default": "true",
+        "Desc": [
+          "When the kernel unaligned atomic handler is enabled, use backpatching to reduce kernel context switches."
+        ]
+      },
       "VolatileMetadata": {
         "Type": "bool",
         "Default": "true",
         "Desc": [
           "Use volatile metadata in PE files to inform TSO instructions when available.",
-	  "When metadata is unavailable falls back to the currently enabled TSO options."
+          "When metadata is unavailable falls back to the currently enabled TSO options."
         ]
       },
       "X87ReducedPrecision": {
@@ -403,23 +465,6 @@
           "Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
         ]
       },
-      "ABILocalFlags": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "When enabled enables an optimization around flags.",
-          "Assumes flags are not used across cals.",
-          "Hand-written assembly can violate this assumption."
-        ]
-      },
-      "ParanoidTSO": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "Makes TSO operations even more strict.",
-          "Forces vector loadstores to also become atomic."
-        ]
-      },
       "StallProcess": {
         "Type": "bool",
         "Default": "false",

FEXCore/Source/Interface/Context/Context.h

@@ -4,7 +4,6 @@
 #include "Common/JitSymbols.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/X86HelperGen.h"
 #include <Interface/IR/IntrusiveIRList.h>
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/Context.h>
@@ -29,6 +28,7 @@
 namespace FEXCore {
 class SignalDelegator;
 class ThunkHandler;
+struct LookupCacheWriteLockToken;
 
 namespace Core {
   struct DebugData;
@@ -61,7 +61,7 @@ struct CustomIRResult {
     , Data(Data) {}
 };
 
-using BlockDelinkerFunc = void (*)(FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record);
+using BlockDelinkerFunc = void (*)(FEXCore::Context::ExitFunctionLinkData* Record);
 constexpr uint32_t TSC_SCALE_MAXIMUM = 1'000'000'000; ///< 1Ghz
 
 class CodeCache : public AbstractCodeCache {
@@ -72,12 +72,34 @@ public:
   ContextImpl& CTX;
   bool IsGeneratingCache = false;
 
+  uint64_t ComputeCodeMapId(std::string_view Filename, int FD) override;
+
   void LoadData(Core::InternalThreadState&, std::byte* MappedCacheFile, const ExecutableFileSectionInfo&) override;
   bool SaveData(Core::InternalThreadState&, int TargetFD, const ExecutableFileSectionInfo&, uint64_t SerializedBaseAddress) override;
 
   void InitiateCacheGeneration() override {
     IsGeneratingCache = true;
   }
+
+  /**
+   * Applies a set of FEX relocations to the given code section.
+   *
+   * FEX relocations describe runtime-dependencies of FEX-generated code.
+   * When loading a code cache, they are used to move cached code to the
+   * dynamically chosen base address of the guest binary.
+   *
+   * Conversely, relocations are applied in reverse when writing code caches
+   * to ensure consistency across generation runs.
+   *
+   * Note that FEX relocations are unrelated to ELF/PE relocations.
+   *
+   * @param GuestDelta Guest address offset to apply to RIP-relative data
+   * @param ForStorage True for serializing data (producing deterministic output); false for de-serializing it (resolving dynamic symbols)
+   *
+   * @return Returns true on success
+   */
+  [[nodiscard]]
+  bool ApplyCodeRelocations(uint64_t GuestDelta, std::span<std::byte> Code, std::span<const CPU::Relocation> Relocations, bool ForStorage);
 };
 
 class ContextImpl final : public FEXCore::Context::Context, public CPU::CodeBufferManager {
@@ -154,10 +176,20 @@ public:
     return CodeCache;
   }
 
-  void OnCodeBufferAllocated(CPU::CodeBuffer&) override;
+  void SetCodeMapWriter(fextl::unique_ptr<CodeMapWriter> Writer) override {
+    CodeMapWriter = std::move(Writer);
+  }
+
+  void FlushAndCloseCodeMap() override {
+    if (CodeMapWriter) {
+      CodeMapWriter.reset();
+    }
+  }
+
+  void OnCodeBufferAllocated(const std::shared_ptr<CPU::CodeBuffer>&) override;
   void ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, bool NewCodeBuffer = true) override;
-  void InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator, uint64_t Start,
-                                uint64_t Length) override;
+  void InvalidateCodeBuffersCodeRange(uint64_t Start, uint64_t Length) override;
+  void InvalidateThreadCachedCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) override;
   FEXCore::ForkableSharedMutex& GetCodeInvalidationMutex() override {
     return CodeInvalidationMutex;
   }
@@ -197,7 +229,6 @@ public:
     FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
     FEX_CONFIG_OPT(VectorTSOEnabled, VECTORTSOENABLED);
     FEX_CONFIG_OPT(MemcpySetTSOEnabled, MEMCPYSETTSOENABLED);
-    FEX_CONFIG_OPT(ABILocalFlags, ABILOCALFLAGS);
     FEX_CONFIG_OPT(SMCChecks, SMCCHECKS);
     FEX_CONFIG_OPT(MaxInstPerBlock, MAXINST);
     FEX_CONFIG_OPT(RootFSPath, ROOTFS);
@@ -205,7 +236,6 @@ public:
     FEX_CONFIG_OPT(LibraryJITNaming, LIBRARYJITNAMING);
     FEX_CONFIG_OPT(BlockJITNaming, BLOCKJITNAMING);
     FEX_CONFIG_OPT(GDBSymbols, GDBSYMBOLS);
-    FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
     FEX_CONFIG_OPT(x87ReducedPrecision, X87REDUCEDPRECISION);
     FEX_CONFIG_OPT(DisableTelemetry, DISABLETELEMETRY);
     FEX_CONFIG_OPT(DisableVixlIndirectCalls, DISABLE_VIXL_INDIRECT_RUNTIME_CALLS);
@@ -226,14 +256,12 @@ public:
   FEXCore::ThunkHandler* ThunkHandler {};
   fextl::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
   CodeCache CodeCache;
+  fextl::unique_ptr<CodeMapWriter> CodeMapWriter;
 
   SignalDelegator* SignalDelegation {};
-  X86GeneratedCode X86CodeGen;
 
   ContextImpl(const FEXCore::HostFeatures& Features);
 
-  static bool ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP);
-
   static void ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame* Frame, uint64_t GuestRIP);
 
   // This is used as a replacement for the SMC writes in the mono callsite backpatcher that avoids atomic operations
@@ -268,9 +296,9 @@ public:
 
   FEXCore::JITSymbols Symbols;
 
-  FEXCore::Utils::PooledAllocatorVirtual OpDispatcherAllocator;
-  FEXCore::Utils::PooledAllocatorVirtual FrontendAllocator;
-  FEXCore::Utils::PooledAllocatorVirtual CPUBackendAllocator;
+  FEXCore::Utils::PooledAllocatorVirtual OpDispatcherAllocator {"FEXMem_OpDispatcher"};
+  FEXCore::Utils::PooledAllocatorVirtual FrontendAllocator {"FEXMem_Frontend"};
+  FEXCore::Utils::PooledAllocatorVirtualWithGuard CPUBackendAllocator {"FEXMem_CPUBackend"};
 
   // If Atomic-based TSO emulation is enabled or not.
   bool IsAtomicTSOEnabled() const {
@@ -311,10 +339,6 @@ protected:
       AtomicTSOEmulationEnabled = false;
       VectorAtomicTSOEmulationEnabled = false;
       MemcpyAtomicTSOEmulationEnabled = false;
-    } else if (Config.ParanoidTSO) {
-      AtomicTSOEmulationEnabled = true;
-      VectorAtomicTSOEmulationEnabled = true;
-      MemcpyAtomicTSOEmulationEnabled = true;
     } else {
       AtomicTSOEmulationEnabled = Config.TSOEnabled;
       VectorAtomicTSOEmulationEnabled = Config.TSOEnabled && Config.VectorTSOEnabled;
@@ -353,5 +377,8 @@ private:
 
   bool MonoDetected = false;
   std::atomic<uint64_t> MonoBackpatcherBlock;
+
+  std::mutex CodeBufferListLock;
+  fextl::vector<std::weak_ptr<CPU::CodeBuffer>> CodeBufferList;
 };
 } // namespace FEXCore::Context

FEXCore/Source/Interface/Core/Addressing.cpp

@@ -7,7 +7,7 @@
 
 namespace FEXCore::IR {
 
-Ref LoadEffectiveAddress(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, bool AddSegmentBase, bool AllowUpperGarbage) {
+Ref LoadEffectiveAddress(IREmitter* IREmit, const AddressMode& A, IR::OpSize GPRSize, bool AddSegmentBase, bool AllowUpperGarbage) {
   Ref Tmp = A.Base;
 
   if (A.Offset) {
@@ -51,8 +51,8 @@ Ref LoadEffectiveAddress(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, b
   return Tmp ?: IREmit->Constant(0);
 }
 
-AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, bool HostSupportsTSOImm9, bool AtomicTSO, bool Vector,
-                              IR::OpSize AccessSize) {
+AddressMode SelectAddressMode(IREmitter* IREmit, const AddressMode& A, IR::OpSize GPRSize, bool HostSupportsTSOImm9, bool AtomicTSO,
+                              bool Vector, IR::OpSize AccessSize) {
   const auto Is32Bit = GPRSize == OpSize::i32Bit;
   const auto GPRSizeMatchesAddrSize = A.AddrSize == GPRSize;
   const auto OffsetIndexToLargeFor32Bit = Is32Bit && (A.Offset <= -16384 || A.Offset >= 16384);
@@ -103,7 +103,7 @@ AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSi
     return {
       .Base = LoadEffectiveAddress(IREmit, B, GPRSize, true /* AddSegmentBase */, false),
       .Index = IREmit->Constant(A.Offset),
-      .IndexType = MEM_OFFSET_SXTX,
+      .IndexType = MemOffsetType::SXTX,
       .IndexScale = 1,
     };
   }
@@ -111,15 +111,17 @@ AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSi
   if (AtomicTSO) {
     // TODO: LRCPC3 support for vector Imm9.
   } else if (!Is32Bit && A.Base && (A.Index || A.Segment) && !A.Offset && (A.IndexScale == 1 || A.IndexScale == AccessSizeAsImm)) {
+    AddressMode B = A;
+
     // ScaledRegisterLoadstore
-    if (A.Index && A.Segment) {
-      A.Base = IREmit->Add(GPRSize, A.Base, A.Segment);
-    } else if (A.Segment) {
-      A.Index = A.Segment;
-      A.IndexScale = 1;
+    if (B.Index && B.Segment) {
+      B.Base = IREmit->Add(GPRSize, B.Base, B.Segment);
+    } else if (B.Segment) {
+      B.Index = B.Segment;
+      B.IndexScale = 1;
     }
 
-    return A;
+    return B;
   }
 
   if (Vector || !AtomicTSO) {
@@ -134,7 +136,7 @@ AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSi
         return {
           .Base = LoadEffectiveAddress(IREmit, B, GPRSize, true /* AddSegmentBase */, false),
           .Index = IREmit->Constant(A.Offset),
-          .IndexType = MEM_OFFSET_SXTX,
+          .IndexType = MemOffsetType::SXTX,
           .IndexScale = 1,
         };
       }

FEXCore/Source/Interface/Core/Addressing.h

@@ -11,17 +11,18 @@ struct AddressMode {
   Ref Segment {nullptr};
   Ref Base {nullptr};
   Ref Index {nullptr};
-  MemOffsetType IndexType = MEM_OFFSET_SXTX;
-  uint8_t IndexScale = 1;
   int64_t Offset = 0;
 
+  MemOffsetType IndexType = MemOffsetType::SXTX;
+  uint8_t IndexScale = 1;
+
   // Size in bytes for the address calculation. 8 for an arm64 hardware mode.
   IR::OpSize AddrSize;
   bool NonTSO;
 };
 
-Ref LoadEffectiveAddress(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, bool AddSegmentBase, bool AllowUpperGarbage = false);
-AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, bool HostSupportsTSOImm9, bool AtomicTSO, bool Vector,
-                              IR::OpSize AccessSize);
+Ref LoadEffectiveAddress(IREmitter* IREmit, const AddressMode& A, IR::OpSize GPRSize, bool AddSegmentBase, bool AllowUpperGarbage = false);
+AddressMode SelectAddressMode(IREmitter* IREmit, const AddressMode& A, IR::OpSize GPRSize, bool HostSupportsTSOImm9, bool AtomicTSO,
+                              bool Vector, IR::OpSize AccessSize);
 
-}; // namespace FEXCore::IR
+} // namespace FEXCore::IR

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

@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
-#include "FEXCore/Core/X86Enums.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Context/Context.h"
 
 #include <FEXCore/Core/CoreState.h>
+#include <FEXCore/Core/X86Enums.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
 

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

@@ -1,30 +1,31 @@
 // SPDX-License-Identifier: MIT
 #pragma once
 
-#include "FEXCore/Utils/EnumUtils.h"
-#include "Interface/Core/JIT/Relocations.h"
-
 #ifdef VIXL_DISASSEMBLER
 #include <aarch64/disasm-aarch64.h>
+#include <FEXCore/Config/Config.h>
+#include <FEXCore/fextl/memory.h>
+#include <FEXCore/fextl/vector.h>
 #endif
 #ifdef VIXL_SIMULATOR
 #include <aarch64/simulator-aarch64.h>
 #include <aarch64/simulator-constants-aarch64.h>
 #endif
 
-#include <FEXCore/Core/X86Enums.h>
-#include <FEXCore/Config/Config.h>
-#include <FEXCore/fextl/vector.h>
 #include <CodeEmitter/Emitter.h>
 #include <CodeEmitter/Registers.h>
 
 #include <cstddef>
 #include <cstdint>
+#include <optional>
 #include <span>
 
 namespace FEXCore::Context {
 class ContextImpl;
 }
+namespace FEXCore::X86State {
+enum X86Reg : uint32_t;
+}
 
 namespace FEXCore::CPU {
 // Contains the address to the currently available CPU state
@@ -104,9 +105,12 @@ constexpr ARMEmitter::PRegister PRED_TMP_32B = ARMEmitter::PReg::p7;
 // This class contains common emitter utility functions that can
 // be used by both Arm64 JIT and ARM64 Dispatcher
 class Arm64Emitter : public ARMEmitter::Emitter {
-protected:
+public:
   Arm64Emitter(FEXCore::Context::ContextImpl* ctx, void* EmissionPtr = nullptr, size_t size = 0);
 
+  void LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad = false);
+
+protected:
   FEXCore::Context::ContextImpl* EmitterCTX;
 
   std::span<const ARMEmitter::Register> StaticRegisters {};
@@ -116,8 +120,6 @@ protected:
   std::span<const ARMEmitter::VRegister> GeneralFPRegisters {};
   uint32_t PairRegisters = 0;
 
-  void LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad = false);
-
   void FillSpecialRegs(ARMEmitter::Register TmpReg, ARMEmitter::Register TmpReg2, bool SetFIZ, bool SetPredRegs);
 
   // Correlate an ARM register back to an x86 register index.

FEXCore/Source/Interface/Core/CPUBackend.cpp

@@ -1,14 +1,17 @@
 // SPDX-License-Identifier: MIT
-#include "FEXCore/IR/IR.h"
-#include "FEXCore/Utils/AllocatorHooks.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUBackend.h"
+#include "Interface/Core/LookupCache.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
+
+#include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/AllocatorHooks.h>
+#include <FEXCore/Utils/PrctlUtils.h>
+
 #include <cstdint>
 
-#include "LookupCache.h"
-
 #ifndef _WIN32
+#include <linux/prctl.h>
 #include <sys/prctl.h>
 #endif
 
@@ -357,6 +360,8 @@ namespace CPU {
       LogMan::Msg::EFmt("Failed to mprotect last page of code buffer.");
     }
 
+    FEXCore::Allocator::VirtualName("FEXMemJIT", reinterpret_cast<void*>(Ptr), Size);
+
     LookupCache = fextl::make_unique<GuestToHostMap>();
   }
 
@@ -395,7 +400,7 @@ namespace CPU {
     Latest = Buffer;
     LatestOffset = 0;
 
-    OnCodeBufferAllocated(*Buffer);
+    OnCodeBufferAllocated(Buffer);
 
     return Buffer;
   }

FEXCore/Source/Interface/Core/CPUBackend.h

@@ -81,7 +81,7 @@ namespace CPU {
     // Protects writes to the latest CodeBuffer and changes to LatestOffset
     FEXCore::ForkableUniqueMutex CodeBufferWriteMutex;
 
-    virtual void OnCodeBufferAllocated(CodeBuffer&) {};
+    virtual void OnCodeBufferAllocated(const std::shared_ptr<CodeBuffer>&) {};
 
   private:
     fextl::shared_ptr<CodeBuffer> Latest;
@@ -161,7 +161,7 @@ namespace CPU {
     virtual CompiledCode CompileCode(uint64_t Entry, uint64_t Size, bool SingleInst, const FEXCore::IR::IRListView* IR,
                                      FEXCore::Core::DebugData* DebugData, bool CheckTF) = 0;
 
-    virtual fextl::vector<FEXCore::CPU::Relocation> TakeRelocations() = 0;
+    virtual fextl::vector<FEXCore::CPU::Relocation> TakeRelocations(uint64_t GuestBaseAddress) = 0;
 
     virtual void ClearCache() {}
 

FEXCore/Source/Interface/Core/CPUID.cpp

@@ -14,6 +14,7 @@ $end_info$
 #include <FEXCore/Core/CPUID.h>
 #include <FEXCore/Core/HostFeatures.h>
 #include <FEXCore/Utils/FileLoading.h>
+#include <FEXCore/Utils/MathUtils.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXHeaderUtils/Syscalls.h>
 
@@ -88,6 +89,7 @@ namespace ProductNames {
   static const char ARM_Blizzard_M2Pro[] = "Apple Blizzard (M2 Pro)";
   static const char ARM_Avalanche_M2Max[] = "Apple Avalanche (M2 Max)";
   static const char ARM_Blizzard_M2Max[] = "Apple Blizzard (M2 Max)";
+  static const char ARM_AppleSilicon[] = "Apple Silicon";
 
   static const char ARM_ORYON_1[] = "Oryon-1";
   static const char ARM_Ampere_1[] = "AmpereOne";
@@ -188,6 +190,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
     {0x61, 0x029, 1, ProductNames::ARM_Firestorm_M1Max}, // Apple Firestorm (M1 Max)
     {0x61, 0x025, 1, ProductNames::ARM_Firestorm_M1Pro}, // Apple Firestorm (M1 Pro)
     {0x61, 0x023, 1, ProductNames::ARM_Firestorm_M1},    // Apple Firestorm (M1)
+    {0x61, 0, 1, ProductNames::ARM_AppleSilicon},        // QEmu Apple Silicon
 
     {0x41, 0xd8c, 1, ProductNames::ARM_C1Ultra},   // C1-Ultra
     {0x41, 0xd90, 1, ProductNames::ARM_C1Premium}, // C1-Premium
@@ -441,10 +444,10 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) const {
 
   Res.eax = FAMILY_IDENTIFIER;
 
-  Res.ebx = 0 |             // Brand index
-            (8 << 8) |      // Cache line size in bytes
-            (Cores << 16) | // Number of addressable IDs for the logical cores in the physical CPU
-            (0 << 24);      // Local APIC ID
+  Res.ebx = 0 |                 // Brand index
+            (8 << 8) |          // Cache line size in bytes
+            (Cores << 16) |     // Number of addressable IDs for the logical cores in the physical CPU
+            (GetCPUID() << 24); // Local APIC ID
 
   Res.ecx = (1 << 0) |                                      // SSE3
             (CTX->HostFeatures.SupportsPMULL_128Bit << 1) | // PCLMULQDQ
@@ -507,7 +510,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) const {
             (1 << 25) | // SSE
             (1 << 26) | // SSE2
             (0 << 27) | // Self Snoop
-            (1 << 28) | // Max APIC IDs reserved field is valid
+            (0 << 28) | // (HTT) Max APIC IDs reserved field is valid
             (1 << 29) | // Thermal monitor
             (0 << 30) | // Reserved
             (0 << 31);  // Pending break enable
@@ -909,38 +912,38 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0001h(uint32_t Leaf) con
 
   Res.eax = FAMILY_IDENTIFIER;
 
-  Res.ecx = (1 << 0) |  // LAHF/SAHF
-            (1 << 1) |  // 0 = Single core product, 1 = multi core product
-            (0 << 2) |  // SVM
-            (1 << 3) |  // Extended APIC register space
-            (0 << 4) |  // LOCK MOV CR0 means MOV CR8
-            (1 << 5) |  // ABM instructions
-            (0 << 6) |  // SSE4a
-            (0 << 7) |  // Misaligned SSE mode
-            (1 << 8) |  // PREFETCHW
-            (0 << 9) |  // OS visible workaround support
-            (0 << 10) | // Instruction based sampling support
-            (0 << 11) | // XOP
-            (0 << 12) | // SKINIT
-            (0 << 13) | // Watchdog timer support
-            (0 << 14) | // Reserved
-            (0 << 15) | // Lightweight profiling support
-            (0 << 16) | // FMA4
-            (1 << 17) | // Translation cache extension
-            (0 << 18) | // Reserved
-            (0 << 19) | // Reserved
-            (0 << 20) | // Reserved
-            (0 << 21) | // XOP-TBM
-            (0 << 22) | // Topology extensions support
-            (0 << 23) | // Core performance counter extensions
-            (0 << 24) | // NB performance counter extensions
-            (0 << 25) | // Reserved
-            (0 << 26) | // Data breakpoints extensions
-            (0 << 27) | // Performance TSC
-            (0 << 28) | // L2 perf counter extensions
-            (0 << 29) | // MONITORX
-            (0 << 30) | // Reserved
-            (0 << 31);  // Reserved
+  Res.ecx = (1 << 0) |                               // LAHF/SAHF
+            (1 << 1) |                               // 0 = Single core product, 1 = multi core product
+            (0 << 2) |                               // SVM
+            (1 << 3) |                               // Extended APIC register space
+            (0 << 4) |                               // LOCK MOV CR0 means MOV CR8
+            (1 << 5) |                               // ABM instructions
+            (CTX->HostFeatures.SupportsSSE4a << 6) | // SSE4a
+            (0 << 7) |                               // Misaligned SSE mode
+            (1 << 8) |                               // PREFETCHW
+            (0 << 9) |                               // OS visible workaround support
+            (0 << 10) |                              // Instruction based sampling support
+            (0 << 11) |                              // XOP
+            (0 << 12) |                              // SKINIT
+            (0 << 13) |                              // Watchdog timer support
+            (0 << 14) |                              // Reserved
+            (0 << 15) |                              // Lightweight profiling support
+            (0 << 16) |                              // FMA4
+            (1 << 17) |                              // Translation cache extension
+            (0 << 18) |                              // Reserved
+            (0 << 19) |                              // Reserved
+            (0 << 20) |                              // Reserved
+            (0 << 21) |                              // XOP-TBM
+            (0 << 22) |                              // Topology extensions support
+            (0 << 23) |                              // Core performance counter extensions
+            (0 << 24) |                              // NB performance counter extensions
+            (0 << 25) |                              // Reserved
+            (0 << 26) |                              // Data breakpoints extensions
+            (0 << 27) |                              // Performance TSC
+            (0 << 28) |                              // L2 perf counter extensions
+            (0 << 29) |                              // MONITORX
+            (0 << 30) |                              // Reserved
+            (0 << 31);                               // Reserved
 
   Res.edx = (1 << 0) |                                // FPU
             (1 << 1) |                                // Virtual mode extensions
@@ -1094,9 +1097,9 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0008h(uint32_t Leaf) con
             (CTX->HostFeatures.SupportsCLZERO << 0); // CLZERO support
 
   uint32_t CoreCount = Cores - 1;
-  Res.ecx = (0 << 16) |                                  // PerfTscSize: Performance timestamp count size
-            ((uint32_t)std::log2(CoreCount + 1) << 12) | // ApicIdSize: Number of bits in ApicID
-            (CoreCount << 0);                            // Count count subtract one
+  Res.ecx = (0 << 16) |                    // PerfTscSize: Performance timestamp count size
+            (std::bit_ceil(Cores) << 12) | // ApicIdSize: Number of bits in ApicID
+            (CoreCount << 0);              // Count count subtract one
 
   return Res;
 }

FEXCore/Source/Interface/Core/CPUID.h

@@ -277,7 +277,7 @@ private:
     // 0: Highest function parameter and ID
     {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
     // 1: Processor info
-    {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
+    {SupportsConstant::NONCONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
     // 2: Cache and TLB info
     {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
     // 3: Serial Number(previously), now reserved

FEXCore/Source/Interface/Core/CodeCache.cpp

@@ -1,12 +1,213 @@
 // SPDX-License-Identifier: MIT
-#include <Interface/Context/Context.h>
+#include "Utils/SpinWaitLock.h"
 
+#include <Interface/Context/Context.h>
+#include <Interface/Core/ArchHelpers/Arm64Emitter.h>
+#include <Interface/Core/JIT/Relocations.h>
+#include <Interface/Core/LookupCache.h>
+
+#include <FEXCore/Core/Thunks.h>
 #include <FEXCore/HLE/SourcecodeResolver.h>
 
+#include <FEXHeaderUtils/Filesystem.h>
+
+#include <git_version.h>
+
+#include <xxhash.h>
+
+#include <fstream>
+
 namespace FEXCore {
 
+#if __clang_major__ < 16
+ExecutableFileInfo::ExecutableFileInfo(fextl::unique_ptr<HLE::SourcecodeMap> Map, uint64_t FileId, fextl::string Filename)
+  : SourcecodeMap(std::move(Map))
+  , FileId(FileId)
+  , Filename(Filename) {}
+#endif
 ExecutableFileInfo::~ExecutableFileInfo() = default;
 
+fextl::string CodeMap::GetBaseFilename(const ExecutableFileInfo& MainExecutable, bool AddNombSuffix) {
+  auto FileId = MainExecutable.FileId;
+
+  std::string_view base_filename = FHU::Filesystem::GetFilename(std::string_view {MainExecutable.Filename});
+  if (FileId != 0xffff'ffff'ffff'ffff) {
+    return fextl::fmt::format("{}-{:016x}{}", base_filename, MainExecutable.FileId, AddNombSuffix ? "-nomb" : "");
+  }
+
+  return "";
+}
+
+fextl::map<CodeMapFileId, CodeMap::ParsedContents> CodeMap::ParseCodeMap(std::ifstream& File) {
+  fextl::map<CodeMapFileId, CodeMap::ParsedContents> Ret;
+  while (true) {
+    Entry Entry;
+    File.read(reinterpret_cast<char*>(&Entry), sizeof(Entry));
+    if (!File) {
+      break;
+    }
+
+    if (Entry.FileId == LoadExternalLibrary.FileId && Entry.BlockOffset == LoadExternalLibrary.BlockOffset) {
+      ExternalLibraryInfo Info;
+      File.read(reinterpret_cast<char*>(&Info), sizeof(Info));
+
+      fextl::string Filename;
+      std::getline(File, Filename, '\0');
+
+      // Align to 4-byte boundary
+      char Null[4];
+      File.read(Null, AlignUp(Filename.size() + 1, 4) - Filename.size() - 1);
+      if (!File) {
+        break;
+      }
+      Ret[Info.ExternalFileId].Filename = std::move(Filename);
+    } else if (Entry.FileId == SetExecutableFileId {}.Marker.FileId && Entry.BlockOffset == SetExecutableFileId {}.Marker.BlockOffset) {
+      CodeMapFileId ExecutableFileId;
+      File.read(reinterpret_cast<char*>(&ExecutableFileId), sizeof(ExecutableFileId));
+      if (!File) {
+        break;
+      }
+      Ret[ExecutableFileId].IsExecutable = true;
+    } else {
+      if (!Ret.contains(Entry.FileId)) {
+        LogMan::Msg::EFmt("Code map referenced unknown file id {:016x}", Entry.FileId);
+      } else {
+        Ret[Entry.FileId].Blocks.insert(Entry.BlockOffset);
+      }
+    }
+
+    if (!File) {
+      break;
+    }
+  }
+  return Ret;
+}
+
+CodeMapWriter::CodeMapWriter(CodeMapOpener& Opener, bool OpenEagerly)
+  : Buffer(4096)
+  , FileOpener(Opener) {
+  if (OpenEagerly) {
+    CodeMapFD = FileOpener.OpenCodeMapFile();
+  }
+}
+
+CodeMapWriter::~CodeMapWriter() {
+  if (CodeMapFD.value_or(-1) != -1) {
+    Flush(BufferOffset);
+    close(*CodeMapFD);
+  }
+}
+
+bool CodeMapWriter::IsWriteEnabled(const ExecutableFileSectionInfo& Section) {
+  if (CodeMapFD == -1) {
+    return false;
+  }
+
+  // PV libraries can't yet be read by FEXServer, so skip dumping them
+  if (Section.FileInfo.Filename.starts_with("/run/pressure-vessel")) {
+    return false;
+  }
+
+  if (CodeMapFD) {
+    return true;
+  }
+
+  // Acquire mutex and re-check CodeMapFD to avoid race conditions
+  auto lk = std::unique_lock {Mutex};
+  if (!CodeMapFD) {
+    CodeMapFD = FileOpener.OpenCodeMapFile();
+  }
+
+  return CodeMapFD != -1;
+}
+
+void CodeMapWriter::Flush(size_t Offset) {
+  // Acquire exclusive lock and flush circular buffer
+  std::unique_lock Lock {Mutex};
+  Flush(Offset, Lock);
+}
+
+void CodeMapWriter::Flush(size_t Offset, std::unique_lock<std::shared_mutex>&) {
+  write(*CodeMapFD, Buffer.data(), Offset);
+  BufferOffset = 0;
+}
+
+void CodeMapWriter::AppendBlock(const FEXCore::ExecutableFileSectionInfo& SectionInfo, uint64_t BlockEntry) {
+  if (!IsWriteEnabled(SectionInfo)) {
+    return;
+  }
+
+  BlockEntry -= SectionInfo.FileStartVA;
+  if (BlockEntry > std::numeric_limits<uint32_t>::max()) {
+    ERROR_AND_DIE_FMT("Cannot write code map");
+  }
+
+  // Register new library if not already known
+  bool NewLibraryLoad = false;
+  {
+    // Check prior registration with shared lock
+    std::shared_lock Lock {Mutex};
+    NewLibraryLoad = !KnownFileIds.contains(SectionInfo.FileInfo.FileId);
+  }
+  if (NewLibraryLoad) {
+    // Register to map with exclusive lock
+    std::unique_lock Lock {Mutex};
+    NewLibraryLoad &= KnownFileIds.insert(SectionInfo.FileInfo.FileId).second;
+  }
+  if (NewLibraryLoad) {
+    // Add entry to code map
+    AppendLibraryLoad(SectionInfo.FileInfo);
+  }
+
+  // Register the actual code block
+  CodeMap::Entry DataEntry {SectionInfo.FileInfo.FileId, static_cast<uint32_t>(BlockEntry)};
+  AppendData(std::as_bytes(std::span {&DataEntry, 1}));
+}
+
+void CodeMapWriter::AppendLibraryLoad(const FEXCore::ExecutableFileInfo& FileInfo) {
+  // See CodeMap::ExternalLibraryInfo
+  auto ExternalFileId = FileInfo.FileId;
+  auto TotalSize = AlignUp(sizeof(CodeMap::LoadExternalLibrary) + sizeof(ExternalFileId) + FileInfo.Filename.size() + 1, 4);
+  const auto Data = reinterpret_cast<char*>(alloca(TotalSize));
+  auto WritePtr = std::copy_n(reinterpret_cast<const char*>(&CodeMap::LoadExternalLibrary), sizeof(CodeMap::LoadExternalLibrary), Data);
+  WritePtr = std::copy_n(reinterpret_cast<const char*>(&ExternalFileId), sizeof(ExternalFileId), WritePtr);
+  WritePtr = std::copy(FileInfo.Filename.begin(), FileInfo.Filename.end(), WritePtr);
+  std::fill(WritePtr, Data + TotalSize, 0);
+  AppendData(std::as_bytes(std::span {Data, TotalSize}));
+}
+
+void CodeMapWriter::AppendSetMainExecutable(const FEXCore::ExecutableFileInfo& FileInfo) {
+  CodeMap::SetExecutableFileId Data {.ExecutableFileId = FileInfo.FileId};
+  AppendData(std::span {reinterpret_cast<const std::byte*>(&Data), sizeof(Data)});
+}
+
+void CodeMapWriter::AppendData(std::span<const std::byte> Data) {
+  std::shared_lock Lock {Mutex};
+  auto Offset = BufferOffset.fetch_add(Data.size_bytes());
+  if (Offset + Data.size_bytes() > Buffer.size()) {
+    // Acquire exclusive lock and flush the buffer.
+    // Under heavy pressure, multiple threads may observe an exhausted buffer simultaneously.
+    // The thread with the last in-bounds Offset is responsible for flushing the buffer.
+    Lock.unlock();
+    bool IsResponsibleForFlush = false;
+    {
+      std::unique_lock ExclusiveLock {Mutex};
+      IsResponsibleForFlush = (Offset <= Buffer.size());
+      if (IsResponsibleForFlush) {
+        Flush(Offset, ExclusiveLock);
+      }
+    }
+    if (!IsResponsibleForFlush) {
+      // Wait for the buffer to be flushed on the responsible thread
+      Utils::SpinWaitLock::WaitPred<std::less_equal<>, size_t>(reinterpret_cast<size_t*>(&BufferOffset), Buffer.size());
+    }
+    AppendData(Data);
+    return;
+  }
+
+  memcpy(&Buffer.at(Offset), Data.data(), Data.size_bytes());
+}
+
 } // namespace FEXCore
 
 namespace FEXCore::Context {
@@ -15,12 +216,156 @@ CodeCache::CodeCache(ContextImpl& CTX_)
   : CTX(CTX_) {}
 CodeCache::~CodeCache() = default;
 
+uint64_t CodeCache::ComputeCodeMapId(std::string_view Filename, int FD) {
+  if (Filename.empty()) {
+    return 0xffff'ffff'ffff'ffff;
+  }
+
+  // For now, we just use the file path as an identifier.
+  // TODO: Ensure the hash is unique enough to distinguish executables while remaining independent of the installation location
+  return XXH3_64bits(Filename.data(), Filename.size());
+}
+
+struct CodeCacheHeader {
+  char Magic[4] = {'F', 'X', 'C', 'C'};
+  uint32_t FormatVersion = 1;
+  char FEXVersion[8] = {};
+  uint32_t NumBlocks;
+  uint32_t NumCodePages;
+  uint32_t CodeBufferSize;
+  uint32_t NumRelocations;
+  uint64_t SerializedBaseAddress;
+  // TODO: Consider including information from LookupCache.BlockLinks
+};
+
 void CodeCache::LoadData(Core::InternalThreadState& Thread, std::byte* MappedCacheFile, const ExecutableFileSectionInfo& GuestRIPLookup) {
   // TODO
 }
 
+template<typename T>
+static constexpr auto IsOrderedContainer(const T&) -> std::false_type;
+template<typename... T>
+static constexpr auto IsOrderedContainer(const std::map<T...>&) -> std::true_type;
+template<typename... T>
+static constexpr auto IsOrderedContainer(const std::set<T...>&) -> std::true_type;
+
 bool CodeCache::SaveData(Core::InternalThreadState& Thread, int fd, const ExecutableFileSectionInfo& SourceBinary, uint64_t SerializedBaseAddress) {
-  // TODO
+  auto CodeBuffer = CTX.GetLatest();
+  auto& LookupCache = *Thread.LookupCache->Shared;
+
+  auto Relocations = Thread.CPUBackend->TakeRelocations(SourceBinary.FileStartVA);
+
+  // Write file header
+  CodeCacheHeader header;
+  memcpy(&header.FEXVersion[0], GIT_SHORT_HASH, strlen(GIT_SHORT_HASH));
+  header.NumBlocks = LookupCache.BlockList.size();
+  header.NumCodePages = LookupCache.CodePages.size();
+  header.CodeBufferSize = CTX.LatestOffset;
+  header.NumRelocations = Relocations.size();
+  header.SerializedBaseAddress = SerializedBaseAddress;
+  ::write(fd, &header, sizeof(header));
+
+  // Dump guest<->host block mappings
+  {
+    // Cache contents must be deterministic, so copy the unordered block list and then sort by key
+    static_assert(!decltype(IsOrderedContainer(LookupCache.BlockList))::value, "Already deterministic; drop temporary container");
+    fextl::vector<std::pair<uint64_t, const GuestToHostMap::BlockEntry*>> BlockList;
+    BlockList.reserve(LookupCache.BlockList.size());
+    for (auto& [Guest, BlockEntry] : LookupCache.BlockList) {
+      static_assert(sizeof(Guest) == 8, "Breaking change in code cache data layout");
+      BlockList.emplace_back(Guest, &BlockEntry);
+    }
+    std::ranges::sort(BlockList);
+
+    for (auto [Guest, Host] : BlockList) {
+      static_assert(sizeof(Host->HostCode) == 8, "Breaking change in code cache data layout");
+      static_assert(sizeof(Host->CodePages[0]) == 8, "Breaking change in code cache data layout");
+
+      Guest -= SourceBinary.FileStartVA;
+      ::write(fd, &Guest, sizeof(Guest));
+      uint64_t HostCode = Host->HostCode - reinterpret_cast<uintptr_t>(CodeBuffer->Ptr);
+      ::write(fd, &HostCode, sizeof(HostCode));
+      uint64_t NumCodePages = Host->CodePages.size();
+      ::write(fd, &NumCodePages, sizeof(NumCodePages));
+      LOGMAN_THROW_A_FMT(std::ranges::is_sorted(Host->CodePages), "Code pages aren't sorted");
+      for (auto CodePage : Host->CodePages) {
+        CodePage -= SourceBinary.FileStartVA;
+        ::write(fd, &CodePage, sizeof(CodePage));
+      }
+    }
+  }
+
+  // Dump relocations
+  static_assert(sizeof(Relocations[0]) == 48, "Breaking change in code cache data layout");
+  ::write(fd, Relocations.data(), Relocations.size() * sizeof(Relocations[0]));
+
+  // Pad to next page in file so that the CodeBuffer can be mmap'ed into process on load
+  char Zero[64] {};
+  auto Off = lseek(fd, 0, SEEK_CUR);
+  while (Off != AlignUp(Off, Utils::FEX_PAGE_SIZE)) {
+    auto BytesToWrite = std::min(AlignUp(Off, Utils::FEX_PAGE_SIZE) - Off, sizeof(Zero));
+    ::write(fd, Zero, BytesToWrite);
+    Off += BytesToWrite;
+  }
+
+  // Dump the host code (relocated for position-independent serialization)
+  std::vector CodeBufferData(reinterpret_cast<std::byte*>(CodeBuffer->Ptr), reinterpret_cast<std::byte*>(CodeBuffer->Ptr) + CTX.LatestOffset);
+  if (!ApplyCodeRelocations(SerializedBaseAddress, CodeBufferData, Relocations, true)) {
+    LOGMAN_THROW_A_FMT(false, "Failed to apply code relocations");
+    return false;
+  }
+  ::write(fd, CodeBufferData.data(), CodeBufferData.size());
+
+  // Dump code pages
+  static_assert(decltype(IsOrderedContainer(LookupCache.CodePages))::value, "Non-deterministic data source");
+  for (auto& [Page, Entrypoints] : LookupCache.CodePages) {
+    static_assert(sizeof(Page) == 8, "Breaking change in code cache data layout");
+    ::write(fd, &Page, sizeof(Page));
+    uint64_t NumEntrypoints = Entrypoints.size();
+    ::write(fd, &NumEntrypoints, sizeof(NumEntrypoints));
+    ::write(fd, Entrypoints.data(), Entrypoints.size() * sizeof(Entrypoints[0]));
+  }
+
+  return true;
+}
+
+bool CodeCache::ApplyCodeRelocations(uint64_t GuestEntry, std::span<std::byte> Code,
+                                     std::span<const FEXCore::CPU::Relocation> EntryRelocations, bool ForStorage) {
+  CPU::Arm64Emitter Emitter(&CTX, Code.data(), Code.size_bytes());
+  for (size_t j = 0; j < EntryRelocations.size(); ++j) {
+    const FEXCore::CPU::Relocation& Reloc = EntryRelocations[j];
+    Emitter.SetCursorOffset(Reloc.Header.Offset);
+
+    switch (Reloc.Header.Type) {
+    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
+      // Generate a literal so we can place it
+      uint64_t Pointer = ForStorage ? 0 : GetNamedSymbolLiteral(CTX, Reloc.NamedSymbolLiteral.Symbol);
+      Emitter.dc64(Pointer);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
+      uint64_t Pointer = ForStorage ? 0 : reinterpret_cast<uint64_t>(CTX.ThunkHandler->LookupThunk(Reloc.NamedThunkMove.Symbol));
+      if (Pointer == ~0ULL) {
+        return false;
+      }
+
+      Emitter.LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.NamedThunkMove.RegisterIndex), Pointer, true);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+      Emitter.dc64(GuestEntry + Reloc.GuestRIP.GuestRIP);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE: {
+      uint64_t Pointer = Reloc.GuestRIP.GuestRIP + GuestEntry;
+      Emitter.LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.GuestRIP.RegisterIndex), Pointer, true);
+      break;
+    }
+
+    default: ERROR_AND_DIE_FMT("Unknown relocation type {}", ToUnderlying(Reloc.Header.Type));
+    }
+  }
+
   return true;
 }
 

FEXCore/Source/Interface/Core/Core.cpp

@@ -376,7 +376,9 @@ void ContextImpl::InitializeCompiler(FEXCore::Core::InternalThreadState* Thread)
   Thread->FrontendDecoder = fextl::make_unique<FEXCore::Frontend::Decoder>(Thread);
   Thread->PassManager = fextl::make_unique<FEXCore::IR::PassManager>();
 
-  Thread->CurrentFrame->Pointers.Common.L1Pointer = Thread->LookupCache->GetL1Pointer();
+  Thread->CurrentFrame->State.L1Pointer = Thread->LookupCache->GetL1Pointer();
+  Thread->CurrentFrame->State.L1Mask = Thread->LookupCache->GetScaledL1PointerMask();
+
   Thread->CurrentFrame->Pointers.Common.L2Pointer = Thread->LookupCache->GetPagePointer();
 
   Dispatcher->InitThreadPointers(Thread);
@@ -398,6 +400,7 @@ ContextImpl::CreateThread(uint64_t InitialRIP, uint64_t StackPointer, const FEXC
   FEXCore::Core::InternalThreadState* Thread = new FEXCore::Core::InternalThreadState {
     .CTX = this,
   };
+  FEXCore::Allocator::VirtualName("FEXMem_ThreadState", Thread, sizeof(*Thread));
 
   Thread->CurrentFrame->State.gregs[X86State::REG_RSP] = StackPointer;
   Thread->CurrentFrame->State.rip = InitialRIP;
@@ -434,6 +437,10 @@ void ContextImpl::UnlockAfterFork(FEXCore::Core::InternalThreadState* LiveThread
 
   Profiler::PostForkAction(Child);
   if (Child) {
+    if (CodeMapWriter) {
+      CodeMapWriter->ResetAfterFork();
+    }
+
     CodeInvalidationMutex.StealAndDropActiveLocks();
     if (Config.StrictInProcessSplitLocks) {
       StrictSplitLockMutex = 0;
@@ -456,9 +463,14 @@ void ContextImpl::LockBeforeFork(FEXCore::Core::InternalThreadState* Thread) {
 }
 #endif
 
-void ContextImpl::OnCodeBufferAllocated(CPU::CodeBuffer& Buffer) {
+void ContextImpl::OnCodeBufferAllocated(const fextl::shared_ptr<CPU::CodeBuffer>& Buffer) {
   if (Config.GlobalJITNaming()) {
-    Symbols.RegisterJITSpace(Buffer.Ptr, Buffer.Size);
+    Symbols.RegisterJITSpace(Buffer->Ptr, Buffer->Size);
+  }
+
+  {
+    std::scoped_lock lk {CodeBufferListLock};
+    CodeBufferList.emplace_back(Buffer);
   }
 }
 
@@ -470,7 +482,8 @@ void ContextImpl::ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, boo
     Thread->CPUBackend->ClearCache();
   } else {
     // Clear L1+L2 cache of this thread, and clear L3 cache across any threads using it
-    Thread->LookupCache->ClearCache();
+    auto lk = Thread->LookupCache->AcquireWriteLock();
+    Thread->LookupCache->ClearCache(lk);
   }
   Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
 }
@@ -642,10 +655,10 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue
               LogMan::Msg::EFmt("Invalid or Unknown instruction: {} 0x{:x}", TableInfo->Name ?: "UND", Block.Entry - GuestRIP);
             }
 
-            if (Block.BlockStatus == Frontend::Decoder::DecodedBlockStatus::NOEXEC_INST) {
-              Thread->OpDispatcher->NoExecOp(DecodedInfo);
-            } else {
+            if (Block.BlockStatus == Frontend::Decoder::DecodedBlockStatus::INVALID_INST) {
               Thread->OpDispatcher->InvalidOp(DecodedInfo);
+            } else {
+              Thread->OpDispatcher->NoExecOp(DecodedInfo);
             }
           }
 
@@ -711,7 +724,8 @@ ContextImpl::CompileCodeResult ContextImpl::CompileCode(FEXCore::Core::InternalT
   if (SourcecodeResolver && Config.GDBSymbols()) {
     auto MappedSection = SyscallHandler->LookupExecutableFileSection(*Thread, GuestRIP);
     if (MappedSection) {
-      MappedSection->FileInfo.SourcecodeMap = SourcecodeResolver->GenerateMap(MappedSection->FileInfo.Filename, MappedSection->FileInfo.FileId);
+      MappedSection->FileInfo.SourcecodeMap =
+        SourcecodeResolver->GenerateMap(MappedSection->FileInfo.Filename, CodeMap::GetBaseFilename(MappedSection->FileInfo, false));
     }
   }
 
@@ -728,7 +742,7 @@ ContextImpl::CompileCodeResult ContextImpl::CompileCode(FEXCore::Core::InternalT
   // but this would increase lock contention. Redundant frontend runs aren't
   // as expensive and are easily reverted.
   if (MaxInst != 1) {
-    if (auto Block = Thread->LookupCache->FindBlock(GuestRIP)) {
+    if (auto Block = Thread->LookupCache->FindBlock(Thread, GuestRIP)) {
       Thread->OpDispatcher->DelayedDisownBuffer();
       return {.CompiledCode = {.BlockBegin = reinterpret_cast<uint8_t*>(Block), .EntryPoints = {{GuestRIP, reinterpret_cast<uint8_t*>(Block)}}},
               .DebugData = nullptr,
@@ -769,10 +783,13 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
 
   // Is the code in the cache?
   // The backends only check L1 and L2, not L3
-  if (auto HostCode = Thread->LookupCache->FindBlock(GuestRIP)) {
+  if (auto HostCode = Thread->LookupCache->FindBlock(Thread, GuestRIP)) {
     return HostCode;
   }
 
+  // Accumulate a JIT count now, as even if another thread raced us, it should count as a compile.
+  FEXCORE_PROFILE_INSTANT_INCREMENT(Thread, AccumulatedJITCount, 1);
+
   auto [CompiledCode, DebugData, StartAddr, Length, NeedsAddGuestCodeRanges] = CompileCode(Thread, GuestRIP, MaxInst);
   auto CodePtr = CompiledCode.EntryPoints[GuestRIP];
   if (CodePtr == nullptr) {
@@ -826,20 +843,32 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
     Thread->CPUBackend->ClearRelocations();
   }
 
+  fextl::vector<uint64_t> CodePages;
+
   if (NeedsAddGuestCodeRanges) {
     // Track in the guest to host map all entrypoints for all pages the compiled block touches, if any page didn't previously
     // contain code, inform the frontend so it can setup SMC detection.
     auto BlockInfo = Thread->FrontendDecoder->GetDecodedBlockInfo();
+    CodePages.reserve(BlockInfo->CodePages.size());
+    CodePages.insert(CodePages.end(), BlockInfo->CodePages.begin(), BlockInfo->CodePages.end());
     for (auto CodePage : BlockInfo->CodePages) {
-      if (Thread->LookupCache->AddBlockExecutableRange(BlockInfo->EntryPoints, CodePage, FEXCore::Utils::FEX_PAGE_SIZE)) {
+      if (Thread->LookupCache->AddBlockExecutableRange(Thread, BlockInfo->EntryPoints, CodePage, FEXCore::Utils::FEX_PAGE_SIZE)) {
         SyscallHandler->MarkGuestExecutableRange(Thread, CodePage, FEXCore::Utils::FEX_PAGE_SIZE);
       }
     }
   }
 
   // Insert to lookup cache
+
   for (auto [GuestAddr, HostAddr] : CompiledCode.EntryPoints) {
-    Thread->LookupCache->AddBlockMapping(GuestAddr, HostAddr);
+    Thread->LookupCache->AddBlockMapping(Thread, GuestAddr, CodePages, HostAddr);
+  }
+
+  if (CodeMapWriter) {
+    auto Region = SyscallHandler->LookupExecutableFileSection(*Thread, GuestRIP);
+    if (Region && Region->FileStartVA != 0) {
+      CodeMapWriter->AppendBlock(*Region, GuestRIP);
+    }
   }
 
   return (uintptr_t)CodePtr;
@@ -866,49 +895,37 @@ uintptr_t ContextImpl::CompileSingleStep(FEXCore::Core::CpuStateFrame* Frame, ui
   return (uintptr_t)CodePtr;
 }
 
-static void InvalidateGuestThreadCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator,
-                                           uint64_t Start, uint64_t Length) {
+void ContextImpl::InvalidateCodeBuffersCodeRange(uint64_t Start, uint64_t Length) {
+  FEXCORE_PROFILE_SCOPED("InvalidateCodeBuffersCodeRange");
+
+  LOGMAN_THROW_A_FMT(CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
+  std::scoped_lock lk {CodeBufferListLock};
+  auto it = CodeBufferList.begin();
+  while (it != CodeBufferList.end()) {
+    if (auto Strong = it->lock()) {
+      Strong->LookupCache->InvalidateRange(Start, Length);
+      it++;
+    } else {
+      it = CodeBufferList.erase(it);
+    }
+  }
+}
+
+void ContextImpl::InvalidateThreadCachedCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) {
+  LOGMAN_THROW_A_FMT(CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
+
   // Ensures now-modified mappings aren't cached as being in their previous non-executable state.
   // Accessing FrontendDecoder is safe as the thread's code invalidation mutex must be locked here.
   Thread->FrontendDecoder->ResetExecutableRangeCache();
 
-  auto lk = Thread->LookupCache->AcquireLock();
-  auto& CodePages = Thread->LookupCache->Shared->CodePages;
+  if (Thread->LookupCache->InvalidateCacheRange(Start, Length)) {
+    FEXCORE_PROFILE_SCOPED("InvalidateCallRet");
 
-  auto lower = CodePages.lower_bound(Start >> 12);
-  auto upper = CodePages.upper_bound((Start + Length - 1) >> 12);
-
-  for (auto it = lower; it != upper; it++) {
-    Accumulator.emplace_back(std::move(it->second));
-  }
-
-  bool InvalidatedAnyEntries = false;
-  for (const auto& PageEntries : Accumulator) {
-    for (const auto& Entry : PageEntries) {
-      if (ContextImpl::ThreadRemoveCodeEntry(Thread, Entry)) {
-        InvalidatedAnyEntries = true;
-      }
-    }
-  }
-
-  if (InvalidatedAnyEntries) {
     // This may cause access violations in the thread on Windows as zeroing is not atomic, this is handled by the frontend
     Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
   }
 }
 
-void ContextImpl::InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator,
-                                           uint64_t Start, uint64_t Length) {
-  InvalidateGuestThreadCodeRange(Thread, Accumulator, Start, Length);
-}
-
-bool ContextImpl::ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP) {
-  LogMan::Throw::AFmt(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to "
-                                                                                                         "be unique_locked here");
-
-  return Thread->LookupCache->Erase(Thread->CurrentFrame, GuestRIP);
-}
-
 void ContextImpl::ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame* Frame, uint64_t GuestRIP) {
   static_cast<ContextImpl*>(Frame->Thread->CTX)->SyscallHandler->InvalidateGuestCodeRange(Frame->Thread, GuestRIP, 1);
 }
@@ -952,10 +969,10 @@ void ContextImpl::AddThunkTrampolineIRHandler(uintptr_t Entrypoint, uintptr_t Gu
 
       if (GPRSize == IR::OpSize::i64Bit) {
         IR::Ref R = emit->_StoreRegister(emit->Constant(Entrypoint), GPRSize);
-        R->Reg = IR::PhysicalRegister(IR::GPRFixedClass, X86State::REG_R11).Raw;
+        R->Reg = IR::PhysicalRegister(IR::RegClass::GPRFixed, X86State::REG_R11).Raw;
       } else {
-        emit->_StoreContext(GPRSize, IR::FPRClass, emit->_VCastFromGPR(IR::OpSize::i64Bit, IR::OpSize::i64Bit, emit->Constant(Entrypoint)),
-                            offsetof(Core::CPUState, mm[0][0]));
+        emit->_StoreContextFPR(GPRSize, emit->_VCastFromGPR(IR::OpSize::i64Bit, IR::OpSize::i64Bit, emit->Constant(Entrypoint)),
+                               offsetof(Core::CPUState, mm[0][0]));
       }
       emit->_ExitFunction(IR::OpSize::i64Bit, emit->Constant(GuestThunkEntrypoint), IR::BranchHint::None, emit->Invalid(), emit->Invalid());
     },
@@ -976,7 +993,7 @@ void ContextImpl::AddThunkTrampolineIRHandler(uintptr_t Entrypoint, uintptr_t Gu
 void ContextImpl::AddForceTSOInformation(const IntervalList<uint64_t>& ValidRanges, fextl::set<uint64_t>&& Instructions) {
   LogMan::Throw::AFmt(CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
   ForceTSOValidRanges.Insert(ValidRanges);
-  ForceTSOInstructions.merge(Instructions);
+  ForceTSOInstructions.merge(std::move(Instructions));
 }
 
 void ContextImpl::RemoveForceTSOInformation(uint64_t Address, uint64_t Size) {

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

@@ -5,7 +5,6 @@
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/LookupCache.h"
-#include "Interface/Core/X86HelperGen.h"
 #include "Utils/MemberFunctionToPointer.h"
 
 #include <FEXCore/Config/Config.h>
@@ -36,12 +35,14 @@ static void SleepThread(FEXCore::Context::ContextImpl* CTX, FEXCore::Core::CpuSt
   CTX->SyscallHandler->SleepThread(CTX, Frame);
 }
 
-constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096 * 4;
+constexpr size_t MAX_DISPATCHER_CODE_SIZE = FEXCore::Utils::FEX_PAGE_SIZE * 4;
 
 Dispatcher::Dispatcher(FEXCore::Context::ContextImpl* ctx)
   : Arm64Emitter(ctx, FEXCore::Allocator::VirtualAlloc(MAX_DISPATCHER_CODE_SIZE, true), MAX_DISPATCHER_CODE_SIZE)
   , CTX {ctx} {
   EmitDispatcher();
+
+  FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(GetBufferBase()), MAX_DISPATCHER_CODE_SIZE);
 }
 
 Dispatcher::~Dispatcher() {
@@ -96,7 +97,7 @@ void Dispatcher::EmitDispatcher() {
   ARMEmitter::BiDirectionalLabel LoopTop {};
 
 #ifdef _M_ARM_64EC
-  b(&LoopTop);
+  (void)b(&LoopTop);
 
   AbsoluteLoopTopAddressEnterECFillSRA = GetCursorAddress<uint64_t>();
   ldr(STATE, EC_ENTRY_CPUAREA_REG, CPU_AREA_EMULATOR_DATA_OFFSET);
@@ -104,10 +105,10 @@ void Dispatcher::EmitDispatcher() {
 
   ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));
   // Force a single instruction block if ENTRY_FILL_SRA_SINGLE_INST_REG is nonzero entering the JIT, used for inline SMC handling.
-  cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
+  (void)cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
 
   // Enter JIT
-  b(&LoopTop);
+  (void)b(&LoopTop);
 
   AbsoluteLoopTopAddressEnterEC = GetCursorAddress<uint64_t>();
   // Load ThreadState and write the target PC there
@@ -128,7 +129,7 @@ void Dispatcher::EmitDispatcher() {
   ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, REG_CALLRET_SP);
   // EC_CALL_CHECKER_PC_REG is REG_PF which isn't touched by any of the above
   sub(ARMEmitter::Size::i64Bit, TMP1, EC_CALL_CHECKER_PC_REG, TMP1);
-  cbnz(ARMEmitter::Size::i64Bit, TMP1, &LoopTop);
+  (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &LoopTop);
 
   // If the entry at the TOS is for the target address, pop it and return to the JIT code
   add(ARMEmitter::Size::i64Bit, REG_CALLRET_SP, REG_CALLRET_SP, 0x10);
@@ -167,67 +168,73 @@ void Dispatcher::EmitDispatcher() {
   ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionEC));
   br(TMP2);
 
-  (void)!Bind(&l_NotECCode);
+  (void)Bind(&l_NotECCode);
 #endif
 
   ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
   (void)cbnz(ARMEmitter::Size::i32Bit, TMP1, &CompileSingleStep);
 
-  // This is the block cache lookup routine
-  // It matches what is going on it LookupCache.h::FindBlock
-  ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
-
-  // Mask the address by the virtual address size so we can check for aliases
-  uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
-  if (std::popcount(VirtualMemorySize) == 1) {
-    and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), VirtualMemorySize - 1);
-  } else {
-    LoadConstant(ARMEmitter::Size::i64Bit, TMP4, VirtualMemorySize);
-    and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), TMP4);
-  }
-
   ARMEmitter::ForwardLabel NoBlock;
 
-  {
-    // Offset the address and add to our page pointer
-    lsr(ARMEmitter::Size::i64Bit, TMP2, TMP4, 12);
+  if (DisableL2Cache()) {
+    (void)b(&NoBlock);
+  } else {
+    // This is the block cache lookup routine
+    // It matches what is going on it LookupCache.h::FindBlock
+    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
 
-    // Load the pointer from the offset
-    ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 3);
+    // Mask the address by the virtual address size so we can check for aliases
+    uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
+    if (std::popcount(VirtualMemorySize) == 1) {
+      and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), VirtualMemorySize - 1);
+    } else {
+      LoadConstant(ARMEmitter::Size::i64Bit, TMP4, VirtualMemorySize);
+      and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), TMP4);
+    }
 
-    // If page pointer is zero then we have no block
-    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &NoBlock);
-
-    // Steal the page offset
-    and_(ARMEmitter::Size::i64Bit, TMP2, TMP4, 0x0FFF);
-
-    // Shift the offset by the size of the block cache entry
-    add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, (int)log2(sizeof(FEXCore::LookupCache::LookupCacheEntry)));
-
-    // The the full LookupCacheEntry with a single LDP.
-    // Check the guest address first to ensure it maps to the address we are currently at.
-    // This fixes aliasing problems
-    ldp<ARMEmitter::IndexType::OFFSET>(TMP4, TMP2, TMP1, 0);
-
-    // If the guest address doesn't match, Compile the block.
-    sub(TMP2, TMP2, RipReg);
-    (void)cbnz(ARMEmitter::Size::i64Bit, TMP2, &NoBlock);
-
-
-    // Check the host address to see if it matches, else compile the block.
-    (void)cbz(ARMEmitter::Size::i64Bit, TMP4, &NoBlock);
-
-    // If we've made it here then we have a real compiled block
     {
-      // update L1 cache
-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
+      // Offset the address and add to our page pointer
+      lsr(ARMEmitter::Size::i64Bit, TMP2, TMP4, 12);
 
-      and_(ARMEmitter::Size::i64Bit, TMP2, RipReg.R(), LookupCache::L1_ENTRIES_MASK);
-      add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, 4);
-      stp<ARMEmitter::IndexType::OFFSET>(TMP4, RipReg, TMP1);
+      // Load the pointer from the offset
+      ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 3);
 
-      // Jump to the block
-      br(TMP4);
+      // If page pointer is zero then we have no block
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &NoBlock);
+
+      // Steal the page offset
+      and_(ARMEmitter::Size::i64Bit, TMP2, TMP4, 0x0FFF);
+
+      // Shift the offset by the size of the block cache entry
+      add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+
+      // The the full LookupCacheEntry with a single LDP.
+      // Check the guest address first to ensure it maps to the address we are currently at.
+      // This fixes aliasing problems
+      ldp<ARMEmitter::IndexType::OFFSET>(TMP4, TMP2, TMP1, 0);
+
+      // If the guest address doesn't match, Compile the block.
+      sub(TMP2, TMP2, RipReg);
+      (void)cbnz(ARMEmitter::Size::i64Bit, TMP2, &NoBlock);
+
+      // Check the host address to see if it matches, else compile the block.
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP4, &NoBlock);
+
+      // If we've made it here then we have a real compiled block
+      {
+        // update L1 cache
+        ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.L1Pointer));
+
+        // Calculate (tmp1 + ((ripreg & L1_ENTRIES_MASK) << 4)) for the address
+        // L1Mask is pre-shifted.
+        and_(ARMEmitter::Size::i64Bit, TMP2, TMP2, RipReg.R(), ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+        add(TMP1, TMP1, TMP2);
+
+        stp<ARMEmitter::IndexType::OFFSET>(TMP4, RipReg, TMP1);
+
+        // Jump to the block
+        br(TMP4);
+      }
     }
   }
 
@@ -481,7 +488,7 @@ void Dispatcher::EmitDispatcher() {
 
     // Now push the callback return trampoline to the guest stack
     // Guest will be misaligned because calling a thunk won't correct the guest's stack once we call the callback from the host
-    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, CTX->X86CodeGen.CallbackReturn);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, CTX->SignalDelegation->GetThunkCallbackRET());
 
     ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
     sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, CTX->Config.Is64BitMode ? 16 : 12);

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

@@ -4,6 +4,7 @@
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 
+#include <FEXCore/Config/Config.h>
 #include <FEXCore/fextl/memory.h>
 
 #include <array>
@@ -50,6 +51,10 @@ public:
   }
 #endif
 
+  uint64_t GetExitFunctionLinkerAddress() const {
+    return ExitFunctionLinkerAddress;
+  }
+
   SignalDelegatorConfig MakeSignalDelegatorConfig() const;
 
 protected:
@@ -92,6 +97,8 @@ private:
 
   void EmitDispatcher();
   uint64_t GenerateABICall(FallbackABI ABI);
+
+  FEX_CONFIG_OPT(DisableL2Cache, DISABLEL2CACHE);
 };
 
 } // namespace FEXCore::CPU

FEXCore/Source/Interface/Core/Frontend.cpp

@@ -9,7 +9,6 @@ $end_info$
 #include "Interface/Context/Context.h"
 #include "Interface/Core/Frontend.h"
 #include "Interface/Core/X86Tables/X86Tables.h"
-#include "Interface/Core/X86HelperGen.h"
 #include "Interface/Core/LookupCache.h"
 
 #include <array>
@@ -90,11 +89,6 @@ Decoder::Decoder(FEXCore::Core::InternalThreadState* Thread)
 }
 
 bool Decoder::CheckRangeExecutable(uint64_t Address, uint64_t Size) {
-  // Treat FEX-internal X86 callbacks as always executable
-  if (EntryPoint == CTX->X86CodeGen.CallbackReturn) {
-    return true;
-  }
-
   while (Address < ExecutableRangeBase || Address + Size > ExecutableRangeEnd) {
     auto RangeInfo = CTX->SyscallHandler->QueryGuestExecutableRange(Thread, Address);
     ExecutableRangeBase = RangeInfo.Base;
@@ -1047,8 +1041,11 @@ Decoder::DecodedBlockStatus Decoder::DecodeInstruction(uint64_t PC) {
     // Put an invalid instruction in the stream so the core can raise SIGILL if hit
     // Error while decoding instruction. We don't know the table or instruction size
     DecodeInst->TableInfo = nullptr;
+    auto Result = ErrorDuringDecoding  ? DecodedBlockStatus::INVALID_INST :
+                  DecodeInst->InstSize ? DecodedBlockStatus::PARTIAL_DECODE_INST :
+                                         DecodedBlockStatus::NOEXEC_INST;
     DecodeInst->InstSize = 0;
-    return ErrorDuringDecoding ? DecodedBlockStatus::INVALID_INST : DecodedBlockStatus::NOEXEC_INST;
+    return Result;
   } else if (!DecodeInst->TableInfo || (DecodeInst->TableInfo->Type == TYPE_INST && !DecodeInst->TableInfo->OpcodeDispatcher.OpDispatch)) {
     // If there wasn't an error during decoding but we have no dispatcher for the instruction then claim invalid instruction.
     return DecodedBlockStatus::INVALID_INST;
@@ -1450,7 +1447,10 @@ void Decoder::DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState* Thre
           EraseBlock = true;
         } else {
           LogMan::Msg::EFmt("{} instruction in entry block: {:X}",
-                            BlockIt->BlockStatus == DecodedBlockStatus::INVALID_INST ? "Invalid" : "NoExec", OpAddress);
+                            BlockIt->BlockStatus == DecodedBlockStatus::INVALID_INST ? "Invalid" :
+                            BlockIt->BlockStatus == DecodedBlockStatus::NOEXEC_INST  ? "NoExec" :
+                                                                                       "PartialDecode",
+                            OpAddress);
         }
         break;
       }

FEXCore/Source/Interface/Core/Frontend.h

@@ -27,6 +27,7 @@ public:
     SUCCESS,
     INVALID_INST,
     NOEXEC_INST,
+    PARTIAL_DECODE_INST,
   };
 
   // New Frontend decoding

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

@@ -50,14 +50,13 @@ DEF_OP(EntrypointOffset) {
   auto Op = IROp->C<IR::IROp_EntrypointOffset>();
 
   auto Constant = Entry + Op->Offset;
-  auto Dst = GetReg(Node);
   uint64_t Mask = ~0ULL;
   const auto OpSize = IROp->Size;
   if (OpSize == IR::OpSize::i32Bit) {
     Mask = 0xFFFF'FFFFULL;
   }
 
-  LoadConstant(ARMEmitter::Size::i64Bit, Dst, Constant & Mask);
+  InsertGuestRIPMove(GetReg(Node), Constant & Mask);
 }
 
 DEF_OP(InlineConstant) {
@@ -372,7 +371,7 @@ DEF_OP(CondSubNZCV) {
 DEF_OP(Neg) {
   auto Op = IROp->C<IR::IROp_Neg>();
 
-  if (Op->Cond == FEXCore::IR::COND_AL) {
+  if (Op->Cond == IR::CondClass::AL) {
     neg(ConvertSize48(IROp), GetReg(Node), GetReg(Op->Src));
   } else {
     cneg(ConvertSize48(IROp), GetReg(Node), GetReg(Op->Src), MapCC(Op->Cond));
@@ -1190,6 +1189,19 @@ DEF_OP(Rev) {
   }
 }
 
+DEF_OP(Rbit) {
+  auto Op = IROp->C<IR::IROp_Rbit>();
+  const auto OpSize = IROp->Size;
+
+  LOGMAN_THROW_A_FMT(OpSize == IR::OpSize::i32Bit || OpSize == IR::OpSize::i64Bit, "Unsupported {} size: {}", __func__, OpSize);
+  const auto EmitSize = ConvertSize48(IROp);
+
+  const auto Dst = GetReg(Node);
+  const auto Src = GetReg(Op->Src);
+
+  rbit(EmitSize, Dst, Src);
+}
+
 DEF_OP(Bfi) {
   auto Op = IROp->C<IR::IROp_Bfi>();
   const auto EmitSize = ConvertSize(IROp);
@@ -1273,6 +1285,16 @@ DEF_OP(Sbfe) {
   sbfx(ConvertSize(IROp), Dst, Src, Op->lsb, Op->Width);
 }
 
+DEF_OP(MaskGenerateFromBitWidth) {
+  auto Op = IROp->C<IR::IROp_MaskGenerateFromBitWidth>();
+  auto BitWidth = GetReg(Op->BitWidth);
+
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP1, -1);
+  cmp(ARMEmitter::Size::i64Bit, BitWidth, 0);
+  lslv(ARMEmitter::Size::i64Bit, TMP2, TMP1, BitWidth);
+  csinv(ARMEmitter::Size::i64Bit, GetReg(Node), TMP1, TMP2, ARMEmitter::Condition::CC_EQ);
+}
+
 DEF_OP(Select) {
   auto Op = IROp->C<IR::IROp_Select>();
   const auto OpSize = IROp->Size;

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

@@ -11,23 +11,18 @@ $end_info$
 #include <FEXCore/Core/Thunks.h>
 
 namespace FEXCore::CPU {
-
-uint64_t Arm64JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+uint64_t GetNamedSymbolLiteral(FEXCore::Context::ContextImpl& CTX, FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
   switch (Op) {
   case FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER:
-    return ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker;
-    break;
-  default: ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op)); break;
+    return CTX.Dispatcher->GetExitFunctionLinkerAddress();
+
+  default: ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
   }
-  return ~0ULL;
 }
 
 void Arm64JITCore::InsertNamedThunkRelocation(ARMEmitter::Register Reg, const IR::SHA256Sum& Sum) {
   Relocation MoveABI {};
-  MoveABI.NamedThunkMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE;
-  // Offset is the offset from the entrypoint of the block
-  auto CurrentCursor = GetCursorAddress<uint8_t*>();
-  MoveABI.NamedThunkMove.Offset = CurrentCursor - CodeData.BlockBegin;
+  MoveABI.NamedThunkMove.Header = {.Offset = GetCursorOffset(), .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE};
   MoveABI.NamedThunkMove.Symbol = Sum;
   MoveABI.NamedThunkMove.RegisterIndex = Reg.Idx();
 
@@ -38,9 +33,9 @@ void Arm64JITCore::InsertNamedThunkRelocation(ARMEmitter::Register Reg, const IR
 }
 
 Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
-  uint64_t Pointer = GetNamedSymbolLiteral(Op);
+  uint64_t Pointer = GetNamedSymbolLiteral(*CTX, Op);
 
-  Arm64JITCore::NamedSymbolLiteralPair Lit {
+  NamedSymbolLiteralPair Lit {
     .Lit = Pointer,
     .MoveABI =
       {
@@ -48,92 +43,72 @@ Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXC
           {
             .Header =
               {
+                .Offset = 0, // Set by PlaceNamedSymbolLiteral
                 .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL,
               },
             .Symbol = Op,
-            .Offset = 0,
           },
       },
   };
   return Lit;
 }
 
-void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair& Lit) {
-  // Offset is the offset from the entrypoint of the block
-  auto CurrentCursor = GetCursorAddress<uint8_t*>();
-  Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - CodeData.BlockBegin;
+void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair Lit) {
+  switch (Lit.MoveABI.Header.Type) {
+  case RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL:
+  case RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+    Lit.MoveABI.Header.Offset = GetCursorOffset();
+    break;
+  }
+
+  default: ERROR_AND_DIE_FMT("Unknown relocation type for {}", __FUNCTION__);
+  }
 
   BindOrRestart(&Lit.Loc);
   dc64(Lit.Lit);
   Relocations.emplace_back(Lit.MoveABI);
 }
 
+auto Arm64JITCore::InsertGuestRIPLiteral(uint64_t GuestRIP) -> NamedSymbolLiteralPair {
+  return {
+    .Lit = GuestRIP,
+    .MoveABI =
+      {
+        .GuestRIP = {.Header =
+                       {
+                         .Offset = 0, // Set by PlaceNamedSymbolLiteral
+                         .Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL,
+                       },
+                     // NOTE: Cache serialization will subtract the guest binary base address later to produce consistency results
+                     .GuestRIP = GuestRIP},
+      },
+  };
+}
+
 void Arm64JITCore::InsertGuestRIPMove(ARMEmitter::Register Reg, uint64_t Constant) {
   Relocation MoveABI {};
-  MoveABI.GuestRIPMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE;
-  // Offset is the offset from the entrypoint of the block
-  auto CurrentCursor = GetCursorAddress<uint8_t*>();
-  MoveABI.GuestRIPMove.Offset = CurrentCursor - CodeData.BlockBegin;
-  MoveABI.GuestRIPMove.GuestRIP = Constant;
-  MoveABI.GuestRIPMove.RegisterIndex = Reg.Idx();
+  MoveABI.GuestRIP.Header = {.Offset = GetCursorOffset(), .Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE};
+  // NOTE: Cache serialization will subtract the guest binary base address later to produce consistency results
+  MoveABI.GuestRIP.GuestRIP = Constant;
+  MoveABI.GuestRIP.RegisterIndex = Reg.Idx();
 
   LoadConstant(ARMEmitter::Size::i64Bit, Reg, Constant, false);
   Relocations.emplace_back(MoveABI);
 }
 
-bool Arm64JITCore::ApplyRelocations(uint64_t GuestEntry, std::span<std::byte> Code, std::span<const FEXCore::CPU::Relocation> Relocations) {
-  const auto OrigBase = GetBufferBase();
-  const auto OrigSize = GetBufferSize();
-  const auto OrigOffset = GetCursorOffset();
-
-  SetBuffer(reinterpret_cast<std::uint8_t*>(Code.data()), Code.size_bytes());
-  for (auto& Reloc : Relocations) {
-    switch (Reloc.Header.Type) {
-    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
-      uint64_t Pointer = GetNamedSymbolLiteral(Reloc.NamedSymbolLiteral.Symbol);
-      // Relocation occurs at the cursorEntry + offset relative to that cursor
-      SetCursorOffset(Reloc.NamedSymbolLiteral.Offset);
-
-      // Generate a literal so we can place it
-      dc64(Pointer);
-      break;
-    }
-    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
-      uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Reloc.NamedThunkMove.Symbol));
-      if (Pointer == ~0ULL) {
-        return false;
-      }
-
-      // Relocation occurs at the cursorEntry + offset relative to that cursor.
-      SetCursorOffset(Reloc.NamedThunkMove.Offset);
-      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.NamedThunkMove.RegisterIndex), Pointer, true);
-      break;
-    }
-    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE: {
-      // XXX: Reenable once the JIT Object Cache is upstream
-      // XXX: Should spin the relocation list, create a list of guest RIP moves, and ask for them all once, reduces lock contention.
-      uint64_t Pointer = ~0ULL; // EmitterCTX->JITObjectCache->FindRelocatedRIP(Reloc->GuestRIPMove.GuestRIP);
-      if (Pointer == ~0ULL) {
-        SetBuffer(OrigBase, OrigSize);
-        SetCursorOffset(OrigOffset);
-        return false;
-      }
-
-      // Relocation occurs at the cursorEntry + offset relative to that cursor.
-      SetCursorOffset(Reloc.GuestRIPMove.Offset);
-      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.GuestRIPMove.RegisterIndex), Pointer, true);
+fextl::vector<FEXCore::CPU::Relocation> Arm64JITCore::TakeRelocations(uint64_t GuestBaseAddress) {
+  // Rebase relocations to library base address
+  for (auto& Relocation : Relocations) {
+    switch (Relocation.Header.Type) {
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE:
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+      Relocation.GuestRIP.GuestRIP -= GuestBaseAddress;
       break;
     }
+    default:;
     }
   }
 
-  SetBuffer(OrigBase, OrigSize);
-  SetCursorOffset(OrigOffset);
-
-  return true;
-}
-
-fextl::vector<FEXCore::CPU::Relocation> Arm64JITCore::TakeRelocations() {
   return std::move(Relocations);
 }
 

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

@@ -138,26 +138,6 @@ DEF_OP(CAS) {
   }
 }
 
-DEF_OP(AtomicXor) {
-  auto Op = IROp->C<IR::IROp_AtomicXor>();
-  const auto EmitSize = ConvertSize(IROp);
-  const auto SubEmitSize = ConvertSubRegSize8(IROp->Size);
-
-  auto MemSrc = GetReg(Op->Addr);
-  auto Src = GetReg(Op->Value);
-
-  if (CTX->HostFeatures.SupportsAtomics) {
-    steorl(SubEmitSize, Src, MemSrc);
-  } else {
-    ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
-    ldaxr(SubEmitSize, TMP2, MemSrc);
-    eor(EmitSize, TMP2, TMP2, Src);
-    stlxr(SubEmitSize, TMP2, TMP2, MemSrc);
-    (void)cbnz(EmitSize, TMP2, &LoopTop);
-  }
-}
-
 DEF_OP(AtomicSwap) {
   auto Op = IROp->C<IR::IROp_AtomicSwap>();
   const auto OpSize = IROp->Size;

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

@@ -174,14 +174,12 @@ DEF_OP(ExitFunction) {
     }
 
     // L1 Cache
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer));
+    ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.L1Pointer));
 
     // Calculate (tmp1 + ((ripreg & L1_ENTRIES_MASK) << 4)) for the address
-    // arithmetic. ubfiz+add is marginally faster on Firestorm than
-    // and+add(shift). Same performance on Cortex.
-    static_assert(LookupCache::L1_ENTRIES_MASK == ((1u << 20) - 1));
-    ubfiz(ARMEmitter::Size::i64Bit, TMP4, RipReg, 4, 20);
-    add(TMP1, TMP1, TMP4);
+    // L1Mask is pre-shifted.
+    and_(ARMEmitter::Size::i64Bit, TMP2, TMP2, RipReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+    add(TMP1, TMP1, TMP2);
 
     ldp<ARMEmitter::IndexType::OFFSET>(TMP2, TMP1, TMP1, 0);
 
@@ -235,16 +233,16 @@ DEF_OP(CondJump) {
     LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1), "CondJump: Expected GPR");
     LOGMAN_THROW_A_FMT(isConst, "CondJump: Expected constant source");
 
-    if (Op->Cond.Val == FEXCore::IR::COND_EQ) {
+    if (Op->Cond == IR::CondClass::EQ) {
       LOGMAN_THROW_A_FMT(Const == 0, "CondJump: Expected 0 source");
       cbz_OrRestart(Size, Reg, TrueTargetLabel);
-    } else if (Op->Cond.Val == FEXCore::IR::COND_NEQ) {
+    } else if (Op->Cond == IR::CondClass::NEQ) {
       LOGMAN_THROW_A_FMT(Const == 0, "CondJump: Expected 0 source");
       cbnz_OrRestart(Size, Reg, TrueTargetLabel);
-    } else if (Op->Cond.Val == FEXCore::IR::COND_TSTZ) {
+    } else if (Op->Cond == IR::CondClass::TSTZ) {
       LOGMAN_THROW_A_FMT(Const < 64, "CondJump: Expected valid bit source");
       tbz_OrRestart(Reg, Const, TrueTargetLabel);
-    } else if (Op->Cond.Val == FEXCore::IR::COND_TSTNZ) {
+    } else if (Op->Cond == IR::CondClass::TSTNZ) {
       LOGMAN_THROW_A_FMT(Const < 64, "CondJump: Expected valid bit source");
       tbnz_OrRestart(Reg, Const, TrueTargetLabel);
     } else {
@@ -262,16 +260,10 @@ DEF_OP(Syscall) {
   // X1: ThreadState
   // X2: Pointer to SyscallArguments
 
-  FEXCore::IR::SyscallFlags Flags = Op->Flags;
   PushDynamicRegs(TMP1);
 
   uint32_t GPRSpillMask = ~0U;
   uint32_t FPRSpillMask = ~0U;
-  if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) == FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) {
-    // Need to spill all caller saved registers still
-    GPRSpillMask = CALLER_GPR_MASK;
-    FPRSpillMask = CALLER_FPR_MASK;
-  }
 
   SpillStaticRegs(TMP1, true, GPRSpillMask, FPRSpillMask);
 
@@ -305,117 +297,22 @@ DEF_OP(Syscall) {
 
   add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);
 
-  if ((Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
-    // Result is now in x0
-    // Fix the stack and any values that were stepped on
-    FillStaticRegs(true, GPRSpillMask, FPRSpillMask, ARMEmitter::Reg::r1, ARMEmitter::Reg::r2);
+  // Result is now in x0
+  // Fix the stack and any values that were stepped on
+  FillStaticRegs(true, GPRSpillMask, FPRSpillMask, ARMEmitter::Reg::r1, ARMEmitter::Reg::r2);
 
-    // 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(ARMEmitter::XReg::zr, 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(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
 
-    PopDynamicRegs();
+  PopDynamicRegs();
 
-    if ((Flags & FEXCore::IR::SyscallFlags::NORETURNEDRESULT) != FEXCore::IR::SyscallFlags::NORETURNEDRESULT) {
-      // Move result to its destination register.
-      // Only if `NORETURNEDRESULT` wasn't set, otherwise we might overwrite the CPUState refilled with `FillStaticRegs`
-      mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
-    }
-  }
-}
+  const auto OSABI = CTX->SyscallHandler->GetOSABI();
 
-DEF_OP(InlineSyscall) {
-  auto Op = IROp->C<IR::IROp_InlineSyscall>();
-  // Arguments are passed as follows:
-  // X8: SyscallNumber - RA INTERSECT
-  // X0: Arg0 & Return
-  // X1: Arg1
-  // X2: Arg2
-  // X3: Arg3
-  // X4: Arg4 - RA INTERSECT
-  // X5: Arg5 - RA INTERSECT
-  // X6: Arg6 - Doesn't exist in x86-64 land. RA INTERSECT
-
-  // One argument is removed from the SyscallArguments::MAX_ARGS since the first argument was syscall number
-  const static std::array<ARMEmitter::XRegister, FEXCore::HLE::SyscallArguments::MAX_ARGS - 1> RegArgs = {
-    {ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::XReg::x2, ARMEmitter::XReg::x3, ARMEmitter::XReg::x4, ARMEmitter::XReg::x5}};
-
-  bool Intersects {};
-  // We always need to spill x8 since we can't know if it is live at this SSA location
-  uint32_t SpillMask = 1U << 8;
-  for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-    if (Op->Header.Args[i].IsInvalid()) {
-      break;
-    }
-
-    auto Reg = GetReg(Op->Header.Args[i]);
-    if (Reg == ARMEmitter::Reg::r8 || Reg == ARMEmitter::Reg::r4 || Reg == ARMEmitter::Reg::r5) {
-
-      SpillMask |= (1U << Reg.Idx());
-      Intersects = true;
-    }
-  }
-
-  // Ordering is incredibly important here
-  // We must spill any overlapping registers first THEN claim we are in a syscall without invalidating state at all
-  // Only spill the registers that intersect with our usage
-  SpillStaticRegs(TMP1, false, SpillMask);
-
-  // Now that we are spilled, store in the state that we are in a syscall
-  // Still without overwriting registers that matter
-  // 16bit LoadConstant to be a single instruction
-  // We must always spill at least one register (x8) so this value always has a bit set
-  // This gives the signal handler a value to check to see if we are in a syscall at all
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SpillMask & 0xFFFF);
-  str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
-
-  // Now that we have claimed to be a syscall we can set up the arguments
-  const auto EmitSize = CTX->Config.Is64BitMode() ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
-  const auto EmitSubSize = CTX->Config.Is64BitMode() ? ARMEmitter::SubRegSize::i64Bit : ARMEmitter::SubRegSize::i32Bit;
-  if (Intersects) {
-    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-      if (Op->Header.Args[i].IsInvalid()) {
-        break;
-      }
-
-      auto Reg = GetReg(Op->Header.Args[i]);
-      if (SpillMask & (1U << Reg.Idx())) {
-        // In the case of intersection with x4, x5, or x8 then these are currently SRA
-        // for registers RAX, RDX, and RSP. Which have just been spilled
-        // Just load back from the context.
-        auto Correlation = GetX86RegRelationToARMReg(Reg);
-        LOGMAN_THROW_A_FMT(Correlation != X86State::REG_INVALID, "Invalid register mapping");
-        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[Correlation]));
-      } else {
-        mov(EmitSize, RegArgs[i].R(), Reg);
-      }
-    }
-  } else {
-    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-      if (Op->Header.Args[i].IsInvalid()) {
-        break;
-      }
-
-      mov(EmitSize, RegArgs[i].R(), GetReg(Op->Header.Args[i]));
-    }
-  }
-
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, Op->HostSyscallNumber);
-  svc(0);
-  // On updated signal mask we can receive a signal RIGHT HERE
-
-  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, ~0U, ARMEmitter::Reg::r8, ARMEmitter::Reg::r1);
-
-    // 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(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
-
-    // Result is now in x0
-    // Move result to its destination register
-    mov(EmitSize, GetReg(Node), ARMEmitter::Reg::r0);
+  if (OSABI != FEXCore::HLE::SyscallOSABI::OS_GENERIC) {
+    // Move result to its destination register.
+    // Only if `NORETURNEDRESULT` wasn't set, otherwise we might overwrite the CPUState refilled with `FillStaticRegs`
+    mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
   }
 }
 
@@ -431,8 +328,7 @@ DEF_OP(Thunk) {
 
   mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetReg(Op->ArgPtr));
 
-  auto thunkFn = static_cast<Context::ContextImpl*>(ThreadState->CTX)->ThunkHandler->LookupThunk(Op->ThunkNameHash);
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, (uintptr_t)thunkFn);
+  InsertNamedThunkRelocation(ARMEmitter::Reg::r2, Op->ThunkNameHash);
   if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
     GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
   } else {

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

@@ -423,11 +423,11 @@ DEF_OP(Vector_FToI) {
     const auto Mask = PRED_TMP_32B.Merging();
 
     switch (Op->Round) {
-    case FEXCore::IR::Round_Nearest.Val: frintn(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Negative_Infinity.Val: frintm(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Positive_Infinity.Val: frintp(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Towards_Zero.Val: frintz(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Host.Val: frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::Nearest: frintn(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::NegInfinity: frintm(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::PosInfinity: frintp(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::TowardsZero: frintz(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::Host: frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
     }
   } else {
     const auto IsScalar = ElementSize == OpSize;
@@ -449,21 +449,21 @@ DEF_OP(Vector_FToI) {
   }
 
       switch (Op->Round) {
-      case IR::Round_Nearest.Val: ROUNDING_FN(frintn); break;
-      case IR::Round_Negative_Infinity.Val: ROUNDING_FN(frintm); break;
-      case IR::Round_Positive_Infinity.Val: ROUNDING_FN(frintp); break;
-      case IR::Round_Towards_Zero.Val: ROUNDING_FN(frintz); break;
-      case IR::Round_Host.Val: ROUNDING_FN(frinti); break;
+      case IR::RoundMode::Nearest: ROUNDING_FN(frintn); break;
+      case IR::RoundMode::NegInfinity: ROUNDING_FN(frintm); break;
+      case IR::RoundMode::PosInfinity: ROUNDING_FN(frintp); break;
+      case IR::RoundMode::TowardsZero: ROUNDING_FN(frintz); break;
+      case IR::RoundMode::Host: ROUNDING_FN(frinti); break;
       }
 
 #undef ROUNDING_FN
     } else {
       switch (Op->Round) {
-      case FEXCore::IR::Round_Nearest.Val: frintn(SubEmitSize, Dst.Q(), Vector.Q()); break;
-      case FEXCore::IR::Round_Negative_Infinity.Val: frintm(SubEmitSize, Dst.Q(), Vector.Q()); break;
-      case FEXCore::IR::Round_Positive_Infinity.Val: frintp(SubEmitSize, Dst.Q(), Vector.Q()); break;
-      case FEXCore::IR::Round_Towards_Zero.Val: frintz(SubEmitSize, Dst.Q(), Vector.Q()); break;
-      case FEXCore::IR::Round_Host.Val: frinti(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::Nearest: frintn(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::NegInfinity: frintm(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::PosInfinity: frintp(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::TowardsZero: frintz(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::Host: frinti(SubEmitSize, Dst.Q(), Vector.Q()); break;
       }
     }
   }
@@ -539,11 +539,11 @@ DEF_OP(Vector_F64ToI32) {
     // Then convert to integers using fcvtzs.
     auto CVTReg = Dst.Z();
     switch (Round) {
-    case IR::Round_Nearest.Val: frintn(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
-    case IR::Round_Negative_Infinity.Val: frintm(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
-    case IR::Round_Positive_Infinity.Val: frintp(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
-    case IR::Round_Towards_Zero.Val: CVTReg = Vector.Z(); break;
-    case IR::Round_Host.Val: frinti(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::Nearest: frintn(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::NegInfinity: frintm(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::PosInfinity: frintp(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::TowardsZero: CVTReg = Vector.Z(); break;
+    case IR::RoundMode::Host: frinti(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
     }
 
     fcvtzs(Dst.Z(), ARMEmitter::SubRegSize::i32Bit, Mask, CVTReg, ARMEmitter::SubRegSize::i64Bit);
@@ -567,11 +567,11 @@ DEF_OP(Vector_F64ToI32) {
 
     ///< Round float to integral depending on rounding mode.
     switch (Round) {
-    case FEXCore::IR::Round_Nearest.Val: frintn(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
-    case FEXCore::IR::Round_Negative_Infinity.Val: frintm(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
-    case FEXCore::IR::Round_Positive_Infinity.Val: frintp(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
-    case FEXCore::IR::Round_Towards_Zero.Val: frintz(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
-    case FEXCore::IR::Round_Host.Val: frinti(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::Nearest: frintn(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::NegInfinity: frintm(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::PosInfinity: frintp(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::TowardsZero: frintz(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::Host: frinti(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
     }
 
     // Now narrow from f64 to f32.

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

@@ -493,7 +493,7 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
   }
 }
 
-static void DirectBlockDelinker(FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record, bool Call) {
+static void DirectBlockDelinker(FEXCore::Context::ExitFunctionLinkData* Record, bool Call) {
   uintptr_t JumpThunkStartAddress = reinterpret_cast<uintptr_t>(Record) - 0x10;
   uintptr_t CallerAddress = JumpThunkStartAddress + Record->CallerOffset;
   auto BranchOffset = JumpThunkStartAddress / 4 - CallerAddress / 4;
@@ -511,11 +511,12 @@ static void DirectBlockDelinker(FEXCore::Core::CpuStateFrame* Frame, FEXCore::Co
   ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(CallerAddress), 4);
 }
 
-static void IndirectBlockDelinker(FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
+static void IndirectBlockDelinker(FEXCore::Context::ExitFunctionLinkData* Record) {
   uintptr_t JumpThunkStartAddress = reinterpret_cast<uintptr_t>(Record) - 0x10;
   uint32_t BranchInst = 0;
   ARMEmitter::Emitter BranchEmit(reinterpret_cast<uint8_t*>(&BranchInst), 4);
-  BranchEmit.b(0x8);
+  // Restore branch +2 instructions to jump to the linker block
+  BranchEmit.b(0x2);
 
   std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(JumpThunkStartAddress)).store(BranchInst, std::memory_order::relaxed);
   ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(JumpThunkStartAddress), 4);
@@ -538,7 +539,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
       // Guard the LookupCache lock with the code invalidation mutex, to avoid issues with forking
       auto lk_inval =
         GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
-      HostCode = Thread->LookupCache->FindBlock(GuestRip);
+      HostCode = Thread->LookupCache->FindBlock(Thread, GuestRip);
     }
     if (!HostCode) {
       // Hold a reference to the code buffer, to avoid linking unmapped code if compilation triggers a recreation.
@@ -563,7 +564,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
   auto lk_inval = GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
 
   // Lock here is necessary to prevent simultaneous linking and delinking
-  auto lk = Thread->LookupCache->AcquireLock();
+  auto lk = Thread->LookupCache->AcquireWriteLock();
 
   // For non-calls, this would extend into the block's code, however that's fine as an out-of-range adr would never
   // be generated avoiding any false positives.
@@ -576,14 +577,12 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
 
     if (KnownCallMarkerInst == ExpectedKnownCallMarkerInst) {
       BranchEmit.bl(BranchOffset);
-      Thread->LookupCache->AddBlockLink(GuestRip, Record, [](FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
-        DirectBlockDelinker(Frame, Record, true);
-      });
+      Thread->LookupCache->AddBlockLink(
+        GuestRip, Record, [](FEXCore::Context::ExitFunctionLinkData* Record) { DirectBlockDelinker(Record, true); }, lk);
     } else {
       BranchEmit.b(BranchOffset);
-      Thread->LookupCache->AddBlockLink(GuestRip, Record, [](FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
-        DirectBlockDelinker(Frame, Record, false);
-      });
+      Thread->LookupCache->AddBlockLink(
+        GuestRip, Record, [](FEXCore::Context::ExitFunctionLinkData* Record) { DirectBlockDelinker(Record, false); }, lk);
     }
 
     std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(CallerAddress)).store(BranchInst, std::memory_order::relaxed);
@@ -602,7 +601,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
     std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(JumpThunkStartAddress)).store(LdrInst, std::memory_order::relaxed);
     ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(JumpThunkStartAddress), 4);
 
-    Thread->LookupCache->AddBlockLink(GuestRip, Record, IndirectBlockDelinker);
+    Thread->LookupCache->AddBlockLink(GuestRip, Record, IndirectBlockDelinker, lk);
   }
 
   return HostCode;
@@ -623,10 +622,10 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::ContextImpl* ctx, FEXCore::Core::In
 
   RAPass = Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA");
 
-  RAPass->AddRegisters(FEXCore::IR::GPRClass, GeneralRegisters.size());
-  RAPass->AddRegisters(FEXCore::IR::GPRFixedClass, StaticRegisters.size());
-  RAPass->AddRegisters(FEXCore::IR::FPRClass, GeneralFPRegisters.size());
-  RAPass->AddRegisters(FEXCore::IR::FPRFixedClass, StaticFPRegisters.size());
+  RAPass->AddRegisters(IR::RegClass::GPR, GeneralRegisters.size());
+  RAPass->AddRegisters(IR::RegClass::GPRFixed, StaticRegisters.size());
+  RAPass->AddRegisters(IR::RegClass::FPR, GeneralFPRegisters.size());
+  RAPass->AddRegisters(IR::RegClass::FPRFixed, StaticFPRegisters.size());
   RAPass->PairRegs = PairRegisters;
 
   {
@@ -678,13 +677,13 @@ void Arm64JITCore::EmitDetectionString() {
 void Arm64JITCore::ClearCache() {
   // NOTE: Holding on to the reference here is required to ensure validity of the WriteLock mutex
   auto PrevCodeBuffer = CurrentCodeBuffer;
-  std::lock_guard lk(PrevCodeBuffer->LookupCache->WriteLock);
+  auto lk = PrevCodeBuffer->LookupCache->AcquireWriteLock();
 
   auto CodeBuffer = GetEmptyCodeBuffer();
   SetBuffer(CodeBuffer->Ptr, CodeBuffer->Size);
   EmitDetectionString();
 
-  ThreadState->LookupCache->ChangeGuestToHostMapping(*PrevCodeBuffer, *CurrentCodeBuffer->LookupCache);
+  ThreadState->LookupCache->ChangeGuestToHostMapping(*PrevCodeBuffer, *CurrentCodeBuffer->LookupCache, lk);
 }
 
 Arm64JITCore::~Arm64JITCore() {}
@@ -784,7 +783,7 @@ void Arm64JITCore::EmitSuspendInterruptCheck() {
 
   ldr(TMP2.W(), STATE_PTR(CpuStateFrame, SuspendDoorbell));
   ARMEmitter::ForwardLabel l_NoSuspend;
-  cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
   brk(SuspendMagic);
   (void)Bind(&l_NoSuspend);
 #endif
@@ -810,22 +809,35 @@ void Arm64JITCore::EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool C
       sub(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::rsp, ARMEmitter::XReg::rsp, TMP1, ARMEmitter::ExtendedType::LSL_64, 0);
     }
   }
+
+  EmitSuspendInterruptCheck();
 }
 
 CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size, bool SingleInst, const FEXCore::IR::IRListView* IR,
                                                    FEXCore::Core::DebugData* DebugData, bool CheckTF) {
   FEXCORE_PROFILE_SCOPED("Arm64::CompileCode");
+
+  const auto PrevNumAllocations = Relocations.size();
+
   this->Entry = Entry;
   this->DebugData = DebugData;
   this->IR = IR;
   RequiresFarARM64Jumps = false;
+  SSANodeMultiplier = 24;
 
-  switch (static_cast<RestartOptions::Control>(FEXCore::LongJump::SetJump(RestartControl.RestartJump))) {
+  // Prepare restart via long jump in case branch encoding fails.
+  // This uses UncheckedLongJump since we don't implement std::longjmp in WoA setups
+  switch (static_cast<RestartOptions::Control>(FEXCore::UncheckedLongJump::SetJump(ThreadState->RestartJump))) {
   case RestartOptions::Control::Incoming:
     // Nothing
     break;
   case RestartOptions::Control::EnableFarARM64Jumps: RequiresFarARM64Jumps = true; break;
-  default: ERROR_AND_DIE_FMT("Unhandled Arm64 restart condition!");
+  case RestartOptions::Control::NeedsLargerJITSpace:
+    // Get rid of the claimed buffer immediately, we can't fit in it at all.
+    TempAllocator.UnclaimBuffer();
+    SSANodeMultiplier *= 2;
+    break;
+  default: LOGMAN_MSG_A_FMT("Unhandled Arm64 restart condition!");
   }
 
   uint32_t SSACount = IR->GetSSACount();
@@ -837,12 +849,19 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
   CodeData.EntryPoints.clear();
 
   // Fairly excessive buffer range to make sure we don't overflow
-  uint32_t BufferRange = 0x1000 + SSACount * 24;
+  // One page baseline, plus SSANodeMultipler bytes, plus another page for guard page.
+  const uint32_t DesiredBufferRange = AlignUp(FEXCore::Utils::FEX_PAGE_SIZE * 2 + SSACount * SSANodeMultiplier, FEXCore::Utils::FEX_PAGE_SIZE);
 
   // JIT output is first written to a temporary buffer and later relocated to the CodeBuffer.
   // This minimizes lock contention of CodeBufferWriteMutex.
-  auto TempCodeBuffer = TempAllocator.ReownOrClaimBuffer(BufferRange);
-  SetBuffer(TempCodeBuffer, BufferRange);
+  auto TempCodeBufferInfo = TempAllocator.ReownOrClaimBufferWithSize(DesiredBufferRange);
+  auto TempCodeBuffer = TempCodeBufferInfo.Ptr;
+  const uint32_t UsableBufferRange = TempCodeBufferInfo.Size - FEXCore::Utils::FEX_PAGE_SIZE;
+
+  SetBuffer(TempCodeBuffer, UsableBufferRange);
+
+  ThreadState->JITGuardPage = reinterpret_cast<uintptr_t>(TempCodeBuffer) + UsableBufferRange;
+  ThreadState->JITGuardOverflowArgument = FEXCore::ToUnderlying(RestartOptions::Control::NeedsLargerJITSpace);
 
   CodeData.BlockBegin = GetCursorAddress<uint8_t*>();
 
@@ -932,8 +951,6 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
     for (auto [CodeNode, IROp] : IR->GetCode(BlockNode)) {
       switch (IROp->Op) {
-#define REGISTER_OP_RT(op, x) \
-  case FEXCore::IR::IROps::OP_##op: std::invoke(RT_##x, this, IROp, CodeNode); break
 #define REGISTER_OP(op, x) \
   case FEXCore::IR::IROps::OP_##op: Op_##x(IROp, CodeNode); break
 
@@ -974,22 +991,28 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
     // This is a ExitFunctionLinkData struct
     BindOrRestart(&l_ExitLink);
-    dc64(0);                                             // HostCode
-    dc64(PendingJumpThunk.GuestRIP);                     // GuestRIP
-    dc64(PendingJumpThunk.CallerAddress - ThunkAddress); // CallerOffset
+    dc64(0);                                                                   // HostCode
+    PlaceNamedSymbolLiteral(InsertGuestRIPLiteral(PendingJumpThunk.GuestRIP)); // GuestRIP
+    dc64(PendingJumpThunk.CallerAddress - ThunkAddress);                       // CallerOffset
   }
 
   BindOrRestart(&l_ExitLink);
-  dc64(ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker);
+  PlaceNamedSymbolLiteral(InsertNamedSymbolLiteral(RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER));
 
   // CodeSize not including the header or tail data.
   const uint64_t CodeOnlySize = GetCursorAddress<uint8_t*>() - CodeBegin;
 
-  // Add the JitCodeTail
+  // Add the JitCodeTail (written later)
   Align(alignof(JITCodeTail));
-  auto JITBlockTailLocation = GetCursorAddress<uint8_t*>();
-  auto JITBlockTail = GetCursorAddress<JITCodeTail*>();
-  CursorIncrement(sizeof(JITCodeTail));
+  const auto JITBlockTailLocation = GetCursorAddress<uint8_t*>();
+  CodeHeader->OffsetToBlockTail = JITBlockTailLocation - CodeData.BlockBegin;
+
+  JITCodeTail JITBlockTail {
+    .RIP = Entry,
+    .GuestSize = Size,
+    .SpinLockFutex = 0,
+    .SingleInst = SingleInst,
+  };
 
   // Entries that live after the JITCodeTail.
   // These entries correlate JIT code regions with guest RIP regions.
@@ -1007,23 +1030,13 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
   //   FEXCore::Utils::vl64 GuestRIPOffset;
   // };
 
-  auto JITRIPEntriesBegin = GetCursorAddress<uint8_t*>();
-
-  // Put the block's RIP entry in the tail.
-  // This will be used for RIP reconstruction in the future.
-  // TODO: This needs to be a data RIP relocation once code caching works.
-  //   Current relocation code doesn't support this feature yet.
-  JITBlockTail->RIP = Entry;
-  JITBlockTail->GuestSize = Size;
-  JITBlockTail->SingleInst = SingleInst;
-  JITBlockTail->SpinLockFutex = 0;
-
+  const auto JITRIPEntriesBegin = JITBlockTailLocation + sizeof(JITBlockTail);
   auto JITRIPEntriesLocation = JITRIPEntriesBegin;
 
   {
     // Store the RIP entries.
-    JITBlockTail->NumberOfRIPEntries = DebugData->GuestOpcodes.size();
-    JITBlockTail->OffsetToRIPEntries = JITRIPEntriesBegin - JITBlockTailLocation;
+    JITBlockTail.NumberOfRIPEntries = DebugData->GuestOpcodes.size();
+    JITBlockTail.OffsetToRIPEntries = JITRIPEntriesBegin - JITBlockTailLocation;
     uintptr_t CurrentRIPOffset = 0;
     uint64_t CurrentPCOffset = 0;
 
@@ -1039,14 +1052,20 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
     }
   }
 
-  CursorIncrement(JITRIPEntriesLocation - JITRIPEntriesBegin);
+  SetCursorOffset(JITRIPEntriesLocation - CodeData.BlockBegin);
   Align();
 
-  CodeHeader->OffsetToBlockTail = JITBlockTailLocation - CodeData.BlockBegin;
-
   CodeData.Size = GetCursorAddress<uint8_t*>() - CodeData.BlockBegin;
 
-  JITBlockTail->Size = CodeData.Size;
+  // Finalize and write block tail data
+  JITBlockTail.Size = CodeData.Size;
+  {
+    auto PrevCur = GetCursorOffset();
+    memcpy(JITBlockTailLocation, &JITBlockTail, sizeof(JITBlockTail));
+    SetCursorOffset(JITBlockTailLocation - CodeData.BlockBegin + offsetof(JITCodeTail, RIP));
+    PlaceNamedSymbolLiteral(InsertGuestRIPLiteral(JITBlockTail.RIP));
+    SetCursorOffset(PrevCur);
+  }
 
   // Migrate the compile output from temporary storage to the actual CodeBuffer.
   // This can block progress in other compiling threads, so the duration of the lock should be as small as possible.
@@ -1055,7 +1074,6 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
     // Query size of generated code
     const auto TempSize = GetCursorOffset();
-    LOGMAN_THROW_A_FMT(TempSize <= BufferRange, "Exceeded bounds of temporary buffer ({:#x} vs {:#x})", TempSize, BufferRange);
 
     // Bring CodeBuffer up to date
     {
@@ -1063,7 +1081,8 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
                                                                                                    "doesn't match up!\n");
       if (auto Prev = CheckCodeBufferUpdate()) {
         Allocator::VirtualDontNeed(ThreadState->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
-        ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache);
+        auto lk = ThreadState->LookupCache->AcquireWriteLock();
+        ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache, lk);
       }
 
       // NOTE: 16-byte alignment of the new cursor offset must be preserved for block linking records
@@ -1086,6 +1105,10 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
     }
     CodeBegin += Delta;
 
+    for (std::size_t Idx = PrevNumAllocations; Idx != Relocations.size(); ++Idx) {
+      Relocations[Idx].Header.Offset += CodeBuffers.LatestOffset;
+    }
+
     // Copy over CodeBuffer contents
     memcpy(GetCursorAddress<uint8_t*>(), TempCodeBuffer, TempSize);
     SetCursorOffset(CodeBuffers.LatestOffset + TempSize);

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

@@ -10,13 +10,17 @@ $end_info$
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
+#include "Interface/Core/JIT/Relocations.h"
 #include "Interface/IR/IR.h"
 #include "Interface/IR/IntrusiveIRList.h"
 #include "Interface/IR/RegisterAllocationData.h"
 
+#include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/fextl/map.h>
+#include <FEXCore/fextl/memory.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXCore/fextl/vector.h>
 #include <FEXCore/Utils/LongJump.h>
@@ -26,16 +30,19 @@ $end_info$
 #include <array>
 #include <cstdint>
 #include <functional>
+#include <optional>
 #include <utility>
 #include <variant>
 
 namespace FEXCore::Core {
 struct InternalThreadState;
 }
-
 namespace FEXCore::Context {
 struct ExitFunctionLinkData;
 }
+namespace FEXCore::IR {
+class RegisterAllocationPass;
+}
 
 namespace FEXCore::CPU {
 class Arm64JITCore final : public CPUBackend, public Arm64Emitter {
@@ -54,9 +61,6 @@ public:
   }
 
 private:
-  FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
-  FEX_CONFIG_OPT(HalfBarrierTSOEnabled, HALFBARRIERTSOENABLED);
-
   const bool HostSupportsSVE128 {};
   const bool HostSupportsSVE256 {};
   const bool HostSupportsAVX256 {};
@@ -64,10 +68,10 @@ private:
   const bool HostSupportsAFP {};
 
   struct RestartOptions {
-    FEXCore::LongJump::JumpBuf RestartJump;
     enum class Control : uint64_t {
       Incoming = 0,
       EnableFarARM64Jumps = 1,
+      NeedsLargerJITSpace = 2,
     };
   };
 
@@ -75,6 +79,8 @@ private:
   // In the rare case when those assumptions are broken, FEX needs to safely restart the JIT.
   RestartOptions RestartControl {};
   bool RequiresFarARM64Jumps {};
+  // Default to 6 instructions per SSA node.
+  uint32_t SSANodeMultiplier {24};
 
   ARMEmitter::BiDirectionalLabel* PendingTargetLabel {};
   ARMEmitter::BiDirectionalLabel* PendingCallReturnTargetLabel {};
@@ -105,11 +111,13 @@ private:
 
   [[nodiscard]]
   ARMEmitter::Register GetReg(IR::PhysicalRegister Reg) const {
-    LOGMAN_THROW_A_FMT(Reg.Class == IR::GPRFixedClass.Val || Reg.Class == IR::GPRClass.Val, "Unexpected Class: {}", Reg.Class);
+    const auto RegClass = Reg.AsRegClass();
 
-    if (Reg.Class == IR::GPRFixedClass.Val) {
+    LOGMAN_THROW_A_FMT(RegClass == IR::RegClass::GPRFixed || RegClass == IR::RegClass::GPR, "Unexpected Class: {}", Reg.Class);
+
+    if (RegClass == IR::RegClass::GPRFixed) {
       return StaticRegisters[Reg.Reg];
-    } else if (Reg.Class == IR::GPRClass.Val) {
+    } else if (RegClass == IR::RegClass::GPR) {
       return GeneralRegisters[Reg.Reg];
     }
 
@@ -128,11 +136,13 @@ private:
 
   [[nodiscard]]
   ARMEmitter::VRegister GetVReg(IR::PhysicalRegister Reg) const {
-    LOGMAN_THROW_A_FMT(Reg.Class == IR::FPRFixedClass.Val || Reg.Class == IR::FPRClass.Val, "Unexpected Class: {}", Reg.Class);
+    const auto RegClass = Reg.AsRegClass();
 
-    if (Reg.Class == IR::FPRFixedClass.Val) {
+    LOGMAN_THROW_A_FMT(RegClass == IR::RegClass::FPRFixed || RegClass == IR::RegClass::FPR, "Unexpected Class: {}", Reg.Class);
+
+    if (RegClass == IR::RegClass::FPRFixed) {
       return StaticFPRegisters[Reg.Reg];
-    } else if (Reg.Class == IR::FPRClass.Val) {
+    } else if (RegClass == IR::RegClass::FPR) {
       return GeneralFPRegisters[Reg.Reg];
     }
 
@@ -150,8 +160,8 @@ private:
   }
 
   [[nodiscard]]
-  FEXCore::IR::RegisterClassType GetRegClass(IR::Ref Node) const {
-    return FEXCore::IR::RegisterClassType {IR::PhysicalRegister(Node).Class};
+  static IR::RegClass GetRegClass(IR::Ref Node) {
+    return IR::PhysicalRegister(Node).AsRegClass();
   }
 
   [[nodiscard]]
@@ -168,7 +178,7 @@ private:
   // Converts IR-base shift type to ARMEmitter shift type.
   // Will be a no-op, only a type conversion since the two definitions match.
   [[nodiscard]]
-  ARMEmitter::ShiftType ConvertIRShiftType(IR::ShiftType Shift) const {
+  static ARMEmitter::ShiftType ConvertIRShiftType(IR::ShiftType Shift) {
     return Shift == IR::ShiftType::LSL ? ARMEmitter::ShiftType::LSL :
            Shift == IR::ShiftType::LSR ? ARMEmitter::ShiftType::LSR :
            Shift == IR::ShiftType::ASR ? ARMEmitter::ShiftType::ASR :
@@ -176,18 +186,23 @@ private:
   }
 
   [[nodiscard]]
-  ARMEmitter::Size ConvertSize(const IR::IROp_Header* Op) {
+  static ARMEmitter::Size ConvertSize(const IR::IROp_Header* Op) {
     return Op->Size == IR::OpSize::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
   }
 
   [[nodiscard]]
-  ARMEmitter::Size ConvertSize48(const IR::IROp_Header* Op) {
+  static ARMEmitter::Size ConvertSize48(const IR::IROp_Header* Op) {
     LOGMAN_THROW_A_FMT(Op->Size == IR::OpSize::i32Bit || Op->Size == IR::OpSize::i64Bit, "Invalid size");
     return ConvertSize(Op);
   }
 
   [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize16(IR::OpSize ElementSize) {
+  static ARMEmitter::Size ConvertSize(IR::OpSize Size) {
+    return Size == IR::OpSize::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+  }
+
+  [[nodiscard]]
+  static ARMEmitter::SubRegSize ConvertSubRegSize16(IR::OpSize ElementSize) {
     LOGMAN_THROW_A_FMT(ElementSize == IR::OpSize::i8Bit || ElementSize == IR::OpSize::i16Bit || ElementSize == IR::OpSize::i32Bit ||
                          ElementSize == IR::OpSize::i64Bit || ElementSize == IR::OpSize::i128Bit,
                        "Invalid size");
@@ -199,105 +214,105 @@ private:
   }
 
   [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize16(const IR::IROp_Header* Op) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize16(const IR::IROp_Header* Op) {
     return ConvertSubRegSize16(Op->ElementSize);
   }
 
   [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize8(IR::OpSize ElementSize) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize8(IR::OpSize ElementSize) {
     LOGMAN_THROW_A_FMT(ElementSize != IR::OpSize::i128Bit, "Invalid size");
     return ConvertSubRegSize16(ElementSize);
   }
 
   [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize8(const IR::IROp_Header* Op) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize8(const IR::IROp_Header* Op) {
     return ConvertSubRegSize8(Op->ElementSize);
   }
 
   [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize4(const IR::IROp_Header* Op) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize4(const IR::IROp_Header* Op) {
     LOGMAN_THROW_A_FMT(Op->ElementSize != IR::OpSize::i64Bit, "Invalid size");
     return ConvertSubRegSize8(Op);
   }
 
   [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize248(const IR::IROp_Header* Op) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize248(const IR::IROp_Header* Op) {
     LOGMAN_THROW_A_FMT(Op->ElementSize != IR::OpSize::i8Bit, "Invalid size");
     return ConvertSubRegSize8(Op);
   }
 
   [[nodiscard]]
-  ARMEmitter::VectorRegSizePair ConvertSubRegSizePair16(const IR::IROp_Header* Op) {
+  static ARMEmitter::VectorRegSizePair ConvertSubRegSizePair16(const IR::IROp_Header* Op) {
     return ARMEmitter::ToVectorSizePair(ConvertSubRegSize16(Op));
   }
 
   [[nodiscard]]
-  ARMEmitter::VectorRegSizePair ConvertSubRegSizePair8(const IR::IROp_Header* Op) {
+  static ARMEmitter::VectorRegSizePair ConvertSubRegSizePair8(const IR::IROp_Header* Op) {
     LOGMAN_THROW_A_FMT(Op->ElementSize != IR::OpSize::i128Bit, "Invalid size");
     return ConvertSubRegSizePair16(Op);
   }
 
   [[nodiscard]]
-  ARMEmitter::VectorRegSizePair ConvertSubRegSizePair248(const IR::IROp_Header* Op) {
+  static ARMEmitter::VectorRegSizePair ConvertSubRegSizePair248(const IR::IROp_Header* Op) {
     LOGMAN_THROW_A_FMT(Op->ElementSize != IR::OpSize::i8Bit, "Invalid size");
     return ConvertSubRegSizePair8(Op);
   }
 
   [[nodiscard]]
-  ARMEmitter::Condition MapCC(IR::CondClassType Cond) {
-    switch (Cond.Val) {
-    case FEXCore::IR::COND_EQ: return ARMEmitter::Condition::CC_EQ;
-    case FEXCore::IR::COND_NEQ: return ARMEmitter::Condition::CC_NE;
-    case FEXCore::IR::COND_SGE: return ARMEmitter::Condition::CC_GE;
-    case FEXCore::IR::COND_SLT: return ARMEmitter::Condition::CC_LT;
-    case FEXCore::IR::COND_SGT: return ARMEmitter::Condition::CC_GT;
-    case FEXCore::IR::COND_SLE: return ARMEmitter::Condition::CC_LE;
-    case FEXCore::IR::COND_UGE: return ARMEmitter::Condition::CC_CS;
-    case FEXCore::IR::COND_ULT: return ARMEmitter::Condition::CC_CC;
-    case FEXCore::IR::COND_UGT: return ARMEmitter::Condition::CC_HI;
-    case FEXCore::IR::COND_ULE: return ARMEmitter::Condition::CC_LS;
-    case FEXCore::IR::COND_FLU: return ARMEmitter::Condition::CC_LT;
-    case FEXCore::IR::COND_FGE: return ARMEmitter::Condition::CC_GE;
-    case FEXCore::IR::COND_FLEU: return ARMEmitter::Condition::CC_LE;
-    case FEXCore::IR::COND_FGT: return ARMEmitter::Condition::CC_GT;
-    case FEXCore::IR::COND_FU:
-    case FEXCore::IR::COND_VS: return ARMEmitter::Condition::CC_VS;
-    case FEXCore::IR::COND_FNU:
-    case FEXCore::IR::COND_VC: return ARMEmitter::Condition::CC_VC;
-    case FEXCore::IR::COND_MI: return ARMEmitter::Condition::CC_MI;
-    case FEXCore::IR::COND_PL: return ARMEmitter::Condition::CC_PL;
+  static ARMEmitter::Condition MapCC(IR::CondClass Cond) {
+    switch (Cond) {
+    case IR::CondClass::EQ: return ARMEmitter::Condition::CC_EQ;
+    case IR::CondClass::NEQ: return ARMEmitter::Condition::CC_NE;
+    case IR::CondClass::SGE: return ARMEmitter::Condition::CC_GE;
+    case IR::CondClass::SLT: return ARMEmitter::Condition::CC_LT;
+    case IR::CondClass::SGT: return ARMEmitter::Condition::CC_GT;
+    case IR::CondClass::SLE: return ARMEmitter::Condition::CC_LE;
+    case IR::CondClass::UGE: return ARMEmitter::Condition::CC_CS;
+    case IR::CondClass::ULT: return ARMEmitter::Condition::CC_CC;
+    case IR::CondClass::UGT: return ARMEmitter::Condition::CC_HI;
+    case IR::CondClass::ULE: return ARMEmitter::Condition::CC_LS;
+    case IR::CondClass::FLU: return ARMEmitter::Condition::CC_LT;
+    case IR::CondClass::FGE: return ARMEmitter::Condition::CC_GE;
+    case IR::CondClass::FLEU: return ARMEmitter::Condition::CC_LE;
+    case IR::CondClass::FGT: return ARMEmitter::Condition::CC_GT;
+    case IR::CondClass::FU:
+    case IR::CondClass::VS: return ARMEmitter::Condition::CC_VS;
+    case IR::CondClass::FNU:
+    case IR::CondClass::VC: return ARMEmitter::Condition::CC_VC;
+    case IR::CondClass::MI: return ARMEmitter::Condition::CC_MI;
+    case IR::CondClass::PL: return ARMEmitter::Condition::CC_PL;
     default: LOGMAN_MSG_A_FMT("Unsupported compare type"); return ARMEmitter::Condition::CC_NV;
     }
   }
 
   [[nodiscard]]
-  bool IsFPR(IR::RegisterClassType Class) const {
-    return Class == IR::FPRClass || Class == IR::FPRFixedClass;
+  static bool IsFPR(IR::RegClass Class) {
+    return Class == IR::RegClass::FPR || Class == IR::RegClass::FPRFixed;
   }
 
   [[nodiscard]]
-  bool IsGPR(IR::RegisterClassType Class) const {
-    return Class == IR::GPRClass || Class == IR::GPRFixedClass;
+  static bool IsGPR(IR::RegClass Class) {
+    return Class == IR::RegClass::GPR || Class == IR::RegClass::GPRFixed;
   }
 
   [[nodiscard]]
-  bool IsGPR(IR::Ref Node) {
+  static bool IsGPR(IR::Ref Node) {
     return IsGPR(GetRegClass(Node));
   }
 
   [[nodiscard]]
-  bool IsFPR(IR::Ref Node) {
+  static bool IsFPR(IR::Ref Node) {
     return IsFPR(GetRegClass(Node));
   }
 
   [[nodiscard]]
-  bool IsGPR(IR::OrderedNodeWrapper Wrap) {
-    return IsGPR(IR::RegisterClassType {IR::PhysicalRegister(Wrap).Class});
+  static bool IsGPR(IR::OrderedNodeWrapper Wrap) {
+    return IsGPR(IR::PhysicalRegister(Wrap).AsRegClass());
   }
 
   [[nodiscard]]
-  bool IsFPR(IR::OrderedNodeWrapper Wrap) {
-    return IsFPR(IR::RegisterClassType {IR::PhysicalRegister(Wrap).Class});
+  static bool IsFPR(IR::OrderedNodeWrapper Wrap) {
+    return IsFPR(IR::PhysicalRegister(Wrap).AsRegClass());
   }
 
   [[nodiscard]]
@@ -339,9 +354,7 @@ private:
   }
 
   // Restart helpers
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void bl_OrRestart(T* Label) {
     if (bl(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
       return;
@@ -349,12 +362,10 @@ private:
 
     // We can support this but currently unnecessary.
     ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void b_OrRestart(T* Label) {
     if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
       return;
@@ -362,12 +373,10 @@ private:
 
     // We can support this but currently unnecessary.
     ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void b_OrRestart(ARMEmitter::Condition Cond, T* Label) {
     if (RequiresFarARM64Jumps) {
       ARMEmitter::ForwardLabel Skip {};
@@ -385,12 +394,10 @@ private:
       return;
     }
 
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void cbz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
     if (RequiresFarARM64Jumps) {
       ARMEmitter::ForwardLabel Skip {};
@@ -408,12 +415,10 @@ private:
       return;
     }
 
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void cbnz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
     if (RequiresFarARM64Jumps) {
       ARMEmitter::ForwardLabel Skip {};
@@ -431,12 +436,10 @@ private:
       return;
     }
 
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void tbz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
     if (RequiresFarARM64Jumps) {
       ARMEmitter::ForwardLabel Skip {};
@@ -454,12 +457,10 @@ private:
       return;
     }
 
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void tbnz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
     if (RequiresFarARM64Jumps) {
       ARMEmitter::ForwardLabel Skip {};
@@ -477,38 +478,40 @@ private:
       return;
     }
 
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void adr_OrRestart(ARMEmitter::Register rd, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      if (LongAddressGen(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Unable to encode long ADR.");
+      }
+      return;
+    }
     if (adr(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
       return;
     }
 
-    // We can support this but currently unnecessary.
-    ERROR_AND_DIE_FMT("Long ADR currently unsupported!");
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void adrp_OrRestart(ARMEmitter::Register rd, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      if (LongAddressGen(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Unable to encode long ADRP.");
+      }
+      return;
+    }
     if (adrp(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
       return;
     }
 
-    // We can support this but currently unnecessary.
-    ERROR_AND_DIE_FMT("Long ADRP currently unsupported!");
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
   void BindOrRestart(T* Label) {
     if (Bind(Label)) {
       return;
@@ -516,11 +519,11 @@ private:
 
     if (RequiresFarARM64Jumps) {
       // This should have been caught before this point.
-      ERROR_AND_DIE_FMT("Oops. Unhandled long bind.");
+      ERROR_AND_DIE_FMT("Unhandled long bind");
       return;
     }
 
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
   }
 
   // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
@@ -533,8 +536,6 @@ private:
    * @name Relocations
    * @{ */
 
-  uint64_t GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
-
   /**
    * @brief A literal pair relocation object for named symbol literals
    */
@@ -571,19 +572,30 @@ private:
    */
   NamedSymbolLiteralPair InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
 
+  /**
+   * @brief Inserts a relocation for a constant value relative to the guest entrypoint
+   *
+   * @param Reg - The GPR to move the guest RIP in to
+   * @param Constant - The guest RIP that will be relocated
+   */
+  NamedSymbolLiteralPair InsertGuestRIPLiteral(uint64_t GuestRIP);
+
   /**
    * @brief Place the named symbol literal relocation in memory
    *
    * @param Lit - Which literal to place
    */
-  void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair& Lit);
+  void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair Lit);
 
   fextl::vector<FEXCore::CPU::Relocation> Relocations;
 
-  ///< Relocation code loading
-  bool ApplyRelocations(uint64_t GuestEntry, std::span<std::byte> Code, std::span<const FEXCore::CPU::Relocation>);
-
-  fextl::vector<FEXCore::CPU::Relocation> TakeRelocations() override;
+  /**
+   * Returns any relocations generated since the last call to TakeRelocations.
+   *
+   * GuestBaseAddress must match the base virtual address to which the
+   * input x86 binary is mapped.
+   */
+  fextl::vector<FEXCore::CPU::Relocation> TakeRelocations(uint64_t GuestBaseAddress) override;
 
   /**  @} */
 
@@ -606,7 +618,7 @@ private:
   void Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize, ARMEmitter::VRegister Dst, ARMEmitter::VRegister IncomingDst,
                            std::optional<ARMEmitter::Register> BaseAddr, ARMEmitter::VRegister VectorIndexLow,
                            std::optional<ARMEmitter::VRegister> VectorIndexHigh, ARMEmitter::VRegister MaskReg, IR::OpSize VectorIndexSize,
-                           size_t DataElementOffsetStart, size_t IndexElementOffsetStart, uint8_t OffsetScale);
+                           size_t DataElementOffsetStart, size_t IndexElementOffsetStart, uint8_t OffsetScale, IR::OpSize AddrSize);
 
   void EmitTFCheck();
 

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

@@ -21,7 +21,7 @@ DEF_OP(LoadContext) {
   const auto Op = IROp->C<IR::IROp_LoadContext>();
   const auto OpSize = IROp->Size;
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     auto Dst = GetReg(Node);
 
     switch (OpSize) {
@@ -52,7 +52,7 @@ DEF_OP(LoadContext) {
 DEF_OP(LoadContextPair) {
   const auto Op = IROp->C<IR::IROp_LoadContextPair>();
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     const auto Dst1 = GetReg(Op->OutValue1);
     const auto Dst2 = GetReg(Op->OutValue2);
 
@@ -78,7 +78,7 @@ DEF_OP(StoreContext) {
   const auto Op = IROp->C<IR::IROp_StoreContext>();
   const auto OpSize = IROp->Size;
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     auto Src = GetZeroableReg(Op->Value);
 
     switch (OpSize) {
@@ -110,7 +110,7 @@ DEF_OP(StoreContextPair) {
   const auto Op = IROp->C<IR::IROp_StoreContextPair>();
   const auto OpSize = IROp->Size;
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     auto Src1 = GetZeroableReg(Op->Value1);
     auto Src2 = GetZeroableReg(Op->Value2);
 
@@ -135,11 +135,11 @@ DEF_OP(StoreContextPair) {
 DEF_OP(LoadRegister) {
   const auto Op = IROp->C<IR::IROp_LoadRegister>();
 
-  if (Op->Class == IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     LOGMAN_THROW_A_FMT(Op->Reg < StaticRegisters.size(), "out of range reg");
 
     mov(GetReg(Node).X(), StaticRegisters[Op->Reg].X());
-  } else if (Op->Class == IR::FPRClass) {
+  } else if (Op->Class == IR::RegClass::FPR) {
     const auto regSize = HostSupportsAVX256 ? IR::OpSize::i256Bit : IR::OpSize::i128Bit;
     LOGMAN_THROW_A_FMT(Op->Reg < StaticFPRegisters.size(), "out of range reg");
     LOGMAN_THROW_A_FMT(IROp->Size == regSize, "expected sized");
@@ -175,12 +175,13 @@ DEF_OP(LoadAF) {
 
 DEF_OP(StoreRegister) {
   const auto Op = IROp->C<IR::IROp_StoreRegister>();
-  auto Reg = IR::PhysicalRegister(Node);
+  const auto Reg = IR::PhysicalRegister(Node);
+  const auto RegClass = Reg.AsRegClass();
 
-  if (Reg.Class == IR::GPRFixedClass) {
+  if (RegClass == IR::RegClass::GPRFixed) {
     // Always use 64-bit, it's faster. Upper bits ignored for 32-bit mode.
     mov(ARMEmitter::Size::i64Bit, GetReg(Reg), GetReg(Op->Value));
-  } else if (Reg.Class == IR::FPRFixedClass) {
+  } else if (RegClass == IR::RegClass::FPRFixed) {
     const auto regSize = HostSupportsAVX256 ? IR::OpSize::i256Bit : IR::OpSize::i128Bit;
     LOGMAN_THROW_A_FMT(IROp->Size == regSize, "expected sized");
 
@@ -193,7 +194,7 @@ DEF_OP(StoreRegister) {
       mov(guest.Q(), host.Q());
     }
   } else {
-    LOGMAN_THROW_A_FMT(false, "Unhandled Op->Class {}", Reg.Class);
+    LOGMAN_THROW_A_FMT(false, "Unhandled Op->Class {}", RegClass);
   }
 }
 
@@ -225,7 +226,7 @@ DEF_OP(LoadContextIndexed) {
 
   const auto Index = GetReg(Op->Index);
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     switch (Op->Stride) {
     case 1:
     case 2:
@@ -288,7 +289,7 @@ DEF_OP(StoreContextIndexed) {
 
   const auto Index = GetReg(Op->Index);
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     const auto Value = GetReg(Op->Value);
 
     switch (Op->Stride) {
@@ -372,7 +373,7 @@ DEF_OP(SpillRegister) {
   const auto OpSize = IROp->Size;
   const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     const auto Src = GetReg(Op->Value);
     switch (OpSize) {
     case IR::OpSize::i8Bit: {
@@ -413,7 +414,7 @@ DEF_OP(SpillRegister) {
     }
     default: LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize); break;
     }
-  } else if (Op->Class == FEXCore::IR::FPRClass) {
+  } else if (Op->Class == FEXCore::IR::RegClass::FPR) {
     const auto Src = GetVReg(Op->Value);
 
     switch (OpSize) {
@@ -452,7 +453,7 @@ DEF_OP(SpillRegister) {
     default: LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize); break;
     }
   } else {
-    LOGMAN_MSG_A_FMT("Unhandled SpillRegister class: {}", Op->Class.Val);
+    LOGMAN_MSG_A_FMT("Unhandled SpillRegister class: {}", Op->Class);
   }
 }
 
@@ -461,7 +462,7 @@ DEF_OP(FillRegister) {
   const auto OpSize = IROp->Size;
   const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     const auto Dst = GetReg(Node);
     switch (OpSize) {
     case IR::OpSize::i8Bit: {
@@ -502,7 +503,7 @@ DEF_OP(FillRegister) {
     }
     default: LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize); break;
     }
-  } else if (Op->Class == FEXCore::IR::FPRClass) {
+  } else if (Op->Class == FEXCore::IR::RegClass::FPR) {
     const auto Dst = GetVReg(Node);
 
     switch (OpSize) {
@@ -541,7 +542,7 @@ DEF_OP(FillRegister) {
     default: LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize); break;
     }
   } else {
-    LOGMAN_MSG_A_FMT("Unhandled FillRegister class: {}", Op->Class.Val);
+    LOGMAN_MSG_A_FMT("Unhandled FillRegister class: {}", Op->Class);
   }
 }
 
@@ -578,14 +579,14 @@ ARMEmitter::ExtendedMemOperand Arm64JITCore::GenerateMemOperand(
       return ARMEmitter::ExtendedMemOperand(Base.X(), ARMEmitter::IndexType::OFFSET, Const);
     } else {
       auto RegOffset = GetReg(Offset);
-      switch (OffsetType.Val) {
-      case IR::MEM_OFFSET_SXTX.Val:
+      switch (OffsetType) {
+      case IR::MemOffsetType::SXTX:
         return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::SXTX, FEXCore::ilog2(OffsetScale));
-      case IR::MEM_OFFSET_UXTW.Val:
+      case IR::MemOffsetType::UXTW:
         return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::UXTW, FEXCore::ilog2(OffsetScale));
-      case IR::MEM_OFFSET_SXTW.Val:
+      case IR::MemOffsetType::SXTW:
         return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
-      default: LOGMAN_MSG_A_FMT("Unhandled GenerateMemOperand OffsetType: {}", OffsetType.Val); break;
+      default: LOGMAN_MSG_A_FMT("Unhandled GenerateMemOperand OffsetType: {}", OffsetType); break;
       }
     }
   }
@@ -612,20 +613,20 @@ ARMEmitter::Register Arm64JITCore::ApplyMemOperand(IR::OpSize AccessSize, ARMEmi
     add(ARMEmitter::Size::i64Bit, Tmp, Base, Tmp, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
   } else {
     auto RegOffset = GetReg(Offset);
-    switch (OffsetType.Val) {
-    case IR::MEM_OFFSET_SXTX.Val:
+    switch (OffsetType) {
+    case IR::MemOffsetType::SXTX:
       add(ARMEmitter::Size::i64Bit, Tmp, Base, RegOffset, ARMEmitter::ExtendedType::SXTX, FEXCore::ilog2(OffsetScale));
       break;
 
-    case IR::MEM_OFFSET_UXTW.Val:
+    case IR::MemOffsetType::UXTW:
       add(ARMEmitter::Size::i64Bit, Tmp, Base, RegOffset, ARMEmitter::ExtendedType::UXTW, FEXCore::ilog2(OffsetScale));
       break;
 
-    case IR::MEM_OFFSET_SXTW.Val:
+    case IR::MemOffsetType::SXTW:
       add(ARMEmitter::Size::i64Bit, Tmp, Base, RegOffset, ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
       break;
 
-    default: LOGMAN_MSG_A_FMT("Unhandled OffsetType: {}", OffsetType.Val); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled OffsetType: {}", OffsetType); break;
     }
   }
   return Tmp;
@@ -676,7 +677,7 @@ ARMEmitter::SVEMemOperand Arm64JITCore::GenerateSVEMemOperand(IR::OpSize AccessS
   // Note that we do nothing with the offset type and offset scale,
   // since SVE loads and stores don't have the ability to perform an
   // optional extension or shift as part of their behavior.
-  LOGMAN_THROW_A_FMT(OffsetType.Val == IR::MEM_OFFSET_SXTX.Val, "Currently only the default offset type (SXTX) is supported.");
+  LOGMAN_THROW_A_FMT(OffsetType == IR::MemOffsetType::SXTX, "Currently only the default offset type (SXTX) is supported.");
 
   const auto RegOffset = GetReg(Offset);
   return ARMEmitter::SVEMemOperand(Base.X(), RegOffset.X());
@@ -689,7 +690,7 @@ DEF_OP(LoadMem) {
   const auto MemReg = GetReg(Op->Addr);
   const auto MemSrc = GenerateMemOperand(OpSize, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     const auto Dst = GetReg(Node);
 
     switch (OpSize) {
@@ -723,7 +724,7 @@ DEF_OP(LoadMemPair) {
   const auto Op = IROp->C<IR::IROp_LoadMemPair>();
   const auto Addr = GetReg(Op->Addr);
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     const auto Dst1 = GetReg(Op->OutValue1);
     const auto Dst2 = GetReg(Op->OutValue2);
 
@@ -751,13 +752,13 @@ DEF_OP(LoadMemTSO) {
 
   const auto MemReg = GetReg(Op->Addr);
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     LOGMAN_THROW_A_FMT(Op->Offset.IsInvalid() || CTX->HostFeatures.SupportsTSOImm9, "unexpected offset");
     LOGMAN_THROW_A_FMT(Op->OffsetScale == 1, "unexpected offset scale");
-    LOGMAN_THROW_A_FMT(Op->OffsetType == IR::MEM_OFFSET_SXTX, "unexpected offset type");
+    LOGMAN_THROW_A_FMT(Op->OffsetType == IR::MemOffsetType::SXTX, "unexpected offset type");
   }
 
-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
     const auto Dst = GetReg(Node);
     uint64_t Offset = 0;
     if (!Op->Offset.IsInvalid()) {
@@ -776,12 +777,10 @@ DEF_OP(LoadMemTSO) {
       case IR::OpSize::i64Bit: ldapur(Dst.X(), MemReg, Offset); break;
       default: LOGMAN_MSG_A_FMT("Unhandled LoadMemTSO size: {}", OpSize); break;
       }
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
     }
-  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == FEXCore::IR::GPRClass) {
+  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == IR::RegClass::GPR) {
     const auto Dst = GetReg(Node);
     if (OpSize == IR::OpSize::i8Bit) {
       // 8bit load is always aligned to natural alignment
@@ -793,12 +792,10 @@ DEF_OP(LoadMemTSO) {
       case IR::OpSize::i64Bit: ldapr(Dst.X(), MemReg); break;
       default: LOGMAN_MSG_A_FMT("Unhandled LoadMemTSO size: {}", OpSize); break;
       }
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
     }
-  } else if (Op->Class == FEXCore::IR::GPRClass) {
+  } else if (Op->Class == IR::RegClass::GPR) {
     const auto Dst = GetReg(Node);
     if (OpSize == IR::OpSize::i8Bit) {
       // 8bit load is always aligned to natural alignment
@@ -810,10 +807,8 @@ DEF_OP(LoadMemTSO) {
       case IR::OpSize::i64Bit: ldar(Dst.X(), MemReg); break;
       default: LOGMAN_MSG_A_FMT("Unhandled LoadMemTSO size: {}", OpSize); break;
       }
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
     }
   } else {
     const auto Dst = GetVReg(Node);
@@ -1045,7 +1040,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
                                        ARMEmitter::VRegister IncomingDst, std::optional<ARMEmitter::Register> BaseAddr,
                                        ARMEmitter::VRegister VectorIndexLow, std::optional<ARMEmitter::VRegister> VectorIndexHigh,
                                        ARMEmitter::VRegister MaskReg, IR::OpSize VectorIndexSize, size_t DataElementOffsetStart,
-                                       size_t IndexElementOffsetStart, uint8_t OffsetScale) {
+                                       size_t IndexElementOffsetStart, uint8_t OffsetScale, IR::OpSize AddrSize) {
   LOGMAN_THROW_A_FMT(ElementSize >= IR::OpSize::i8Bit && ElementSize <= IR::OpSize::i64Bit, "Invalid element size");
 
   const auto PerformSMove = [this](IR::OpSize ElementSize, const ARMEmitter::Register Dst, const ARMEmitter::VRegister Vector, int index) {
@@ -1121,17 +1116,17 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
     // Calculate memory position for this gather load
     if (BaseAddr.has_value()) {
       if (VectorIndexSize == IR::OpSize::i32Bit) {
-        add(ARMEmitter::Size::i64Bit, TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
+        add(ConvertSize(AddrSize), TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
       } else {
-        add(ARMEmitter::Size::i64Bit, TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
+        add(ConvertSize(AddrSize), TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
       }
     } else {
       ///< In this case we have no base address, All addresses come from the vector register itself
       if (VectorIndexSize == IR::OpSize::i32Bit) {
         // Sign extend and shift in to the 64-bit register
-        sbfiz(ARMEmitter::Size::i64Bit, TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale), 32);
+        sbfiz(ConvertSize(AddrSize), TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale), 32);
       } else {
-        lsl(ARMEmitter::Size::i64Bit, TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale));
+        lsl(ConvertSize(AddrSize), TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale));
       }
     }
 
@@ -1189,7 +1184,8 @@ DEF_OP(VLoadVectorGatherMasked) {
 
   ///< If the host supports SVE and the offset scale matches SVE limitations then it can do an SVE style load.
   const bool SupportsSVELoad = (HostSupportsSVE128 || HostSupportsSVE256) &&
-                               (OffsetScale == 1 || OffsetScale == IR::OpSizeToSize(VectorIndexSize)) && VectorIndexSize == IROp->ElementSize;
+                               (OffsetScale == 1 || OffsetScale == IR::OpSizeToSize(VectorIndexSize)) &&
+                               VectorIndexSize == IROp->ElementSize && Op->AddrSize == IR::OpSize::i64Bit;
 
   if (SupportsSVELoad) {
     uint8_t SVEScale = FEXCore::ilog2(OffsetScale);
@@ -1247,7 +1243,7 @@ DEF_OP(VLoadVectorGatherMasked) {
   } else {
     LOGMAN_THROW_A_FMT(!Is256Bit, "Can't emulate this gather load in the backend! Programming error!");
     Emulate128BitGather(IROp->Size, IROp->ElementSize, Dst, IncomingDst, BaseAddr, VectorIndexLow, VectorIndexHigh, MaskReg,
-                        VectorIndexSize, DataElementOffsetStart, IndexElementOffsetStart, OffsetScale);
+                        VectorIndexSize, DataElementOffsetStart, IndexElementOffsetStart, OffsetScale, Op->AddrSize);
   }
 }
 
@@ -1272,7 +1268,9 @@ DEF_OP(VLoadVectorGatherMaskedQPS) {
     !Op->VectorIndexHigh.IsInvalid() ? std::make_optional(GetVReg(Op->VectorIndexHigh)) : std::nullopt;
 
   ///< If the host supports SVE and the offset scale matches SVE limitations then it can do an SVE style load.
-  if (HostSupportsSVE128 && (OffsetScale == 1 || OffsetScale == 4)) {
+  const bool SupportsSVELoad = HostSupportsSVE128 && (OffsetScale == 1 || OffsetScale == 4) && Op->AddrSize == IR::OpSize::i64Bit;
+
+  if (SupportsSVELoad) {
     ARMEmitter::SVEModType ModType = ARMEmitter::SVEModType::MOD_NONE;
     if (OffsetScale != 1) {
       ModType = ARMEmitter::SVEModType::MOD_LSL;
@@ -1326,7 +1324,7 @@ DEF_OP(VLoadVectorGatherMaskedQPS) {
     }
   } else {
     Emulate128BitGather(IR::OpSize::i128Bit, IR::OpSize::i32Bit, Dst, IncomingDst, BaseAddr, VectorIndexLow, VectorIndexHigh, MaskReg,
-                        IR::OpSize::i64Bit, 0, 0, OffsetScale);
+                        IR::OpSize::i64Bit, 0, 0, OffsetScale, Op->AddrSize);
   }
 }
 
@@ -1627,7 +1625,7 @@ DEF_OP(StoreMem) {
   const auto MemReg = GetReg(Op->Addr);
   const auto MemSrc = GenerateMemOperand(OpSize, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     const auto Src = GetZeroableReg(Op->Value);
     switch (OpSize) {
     case IR::OpSize::i8Bit: strb(Src, MemSrc); break;
@@ -1738,7 +1736,7 @@ DEF_OP(StoreMemPair) {
   const auto OpSize = IROp->Size;
   const auto Addr = GetReg(Op->Addr);
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     const auto Src1 = GetZeroableReg(Op->Value1);
     const auto Src2 = GetZeroableReg(Op->Value2);
     switch (OpSize) {
@@ -1765,13 +1763,13 @@ DEF_OP(StoreMemTSO) {
 
   const auto MemReg = GetReg(Op->Addr);
 
-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
     LOGMAN_THROW_A_FMT(Op->Offset.IsInvalid() || CTX->HostFeatures.SupportsTSOImm9, "unexpected offset");
     LOGMAN_THROW_A_FMT(Op->OffsetScale == 1, "unexpected offset scale");
-    LOGMAN_THROW_A_FMT(Op->OffsetType == IR::MEM_OFFSET_SXTX, "unexpected offset type");
+    LOGMAN_THROW_A_FMT(Op->OffsetType == IR::MemOffsetType::SXTX, "unexpected offset type");
   }
 
-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
     const auto Src = GetZeroableReg(Op->Value);
     uint64_t Offset = 0;
     if (!Op->Offset.IsInvalid()) {
@@ -1783,10 +1781,8 @@ DEF_OP(StoreMemTSO) {
       // 8bit load is always aligned to natural alignment
       stlurb(Src, MemReg, Offset);
     } else {
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
       switch (OpSize) {
       case IR::OpSize::i16Bit: stlurh(Src, MemReg, Offset); break;
       case IR::OpSize::i32Bit: stlur(Src.W(), MemReg, Offset); break;
@@ -1794,17 +1790,15 @@ DEF_OP(StoreMemTSO) {
       default: LOGMAN_MSG_A_FMT("Unhandled StoreMemTSO size: {}", OpSize); break;
       }
     }
-  } else if (Op->Class == FEXCore::IR::GPRClass) {
+  } else if (Op->Class == IR::RegClass::GPR) {
     const auto Src = GetZeroableReg(Op->Value);
 
     if (OpSize == IR::OpSize::i8Bit) {
       // 8bit load is always aligned to natural alignment
       stlrb(Src, MemReg);
     } else {
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
       switch (OpSize) {
       case IR::OpSize::i16Bit: stlrh(Src, MemReg); break;
       case IR::OpSize::i32Bit: stlr(Src.W(), MemReg); break;
@@ -1898,9 +1892,7 @@ DEF_OP(MemSet) {
       // 8bit load is always aligned to natural alignment
       stlrb(Value.W(), TMP2);
     } else {
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        nop();
-      }
+      nop();
       switch (OpSize) {
       case 2: stlrh(Value.W(), TMP2); break;
       case 4: stlr(Value.W(), TMP2); break;
@@ -2118,11 +2110,9 @@ DEF_OP(MemCpy) {
         default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size); break;
         }
 
-        if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-          // Placeholders for backpatching barriers (one per load/store)
-          nop();
-          nop();
-        }
+        // Placeholders for backpatching barriers (one per load/store)
+        nop();
+        nop();
 
         switch (OpSize) {
         case 2: stlrh(TMP4.W(), TMP2); break;
@@ -2144,11 +2134,9 @@ DEF_OP(MemCpy) {
         default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size); break;
         }
 
-        if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-          // Placeholders for backpatching barriers (one per load/store)
-          nop();
-          nop();
-        }
+        // Placeholders for backpatching barriers (one per load/store)
+        nop();
+        nop();
 
         switch (OpSize) {
         case 2: stlrh(TMP4.W(), TMP2); break;

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

@@ -15,6 +15,7 @@ $end_info$
 
 #include <FEXCore/Core/SignalDelegator.h>
 #include <FEXCore/Debug/InternalThreadState.h>
+#include <FEXCore/Utils/EnumUtils.h>
 
 namespace FEXCore::CPU {
 
@@ -47,10 +48,10 @@ DEF_OP(GuestOpcode) {
 DEF_OP(Fence) {
   auto Op = IROp->C<IR::IROp_Fence>();
   switch (Op->Fence) {
-  case IR::Fence_Load.Val: dmb(ARMEmitter::BarrierScope::LD); break;
-  case IR::Fence_LoadStore.Val: dmb(ARMEmitter::BarrierScope::SY); break;
-  case IR::Fence_Store.Val: dmb(ARMEmitter::BarrierScope::ST); break;
-  case IR::Fence_Inst.Val: isb(); break;
+  case IR::FenceType::Load: dmb(ARMEmitter::BarrierScope::LD); break;
+  case IR::FenceType::LoadStore: dmb(ARMEmitter::BarrierScope::SY); break;
+  case IR::FenceType::Store: dmb(ARMEmitter::BarrierScope::ST); break;
+  case IR::FenceType::Inst: isb(); break;
   default: LOGMAN_MSG_A_FMT("Unknown Fence: {}", Op->Fence); break;
   }
 }
@@ -107,10 +108,10 @@ DEF_OP(GetRoundingMode) {
   // zero. Just swapping 01 and 10. That's a bitfield reverse. Round mode is in
   // bottom two bits. After reversing as a 32-bit operation, it'll be in [31:30]
   // and ripe for reinsertion back at 0.
-  static_assert(IR::ROUND_MODE_NEAREST == 0);
-  static_assert(IR::ROUND_MODE_NEGATIVE_INFINITY == 1);
-  static_assert(IR::ROUND_MODE_POSITIVE_INFINITY == 2);
-  static_assert(IR::ROUND_MODE_TOWARDS_ZERO == 3);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::Nearest) == 0);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::NegInfinity) == 1);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::PosInfinity) == 2);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::TowardsZero) == 3);
 
   rbit(ARMEmitter::Size::i32Bit, TMP1, Dst);
   bfi(ARMEmitter::Size::i64Bit, Dst, TMP1, 30, 2);

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

@@ -18,18 +18,4 @@ DEF_OP(RMWHandle) {
   mov(ARMEmitter::Size::i64Bit, GetReg(Node), GetReg(IROp->Args[0]));
 }
 
-DEF_OP(Swap1) {
-  auto Op = IROp->C<IR::IROp_Swap1>();
-  auto A = GetReg(Op->A), B = GetReg(Op->B);
-  LOGMAN_THROW_A_FMT(B == GetReg(Node), "Invariant");
-
-  mov(ARMEmitter::Size::i64Bit, TMP1, A);
-  mov(ARMEmitter::Size::i64Bit, A, B);
-  mov(ARMEmitter::Size::i64Bit, B, TMP1);
-}
-
-DEF_OP(Swap2) {
-  // Implemented above
-}
-
 } // namespace FEXCore::CPU

FEXCore/Source/Interface/Core/JIT/Relocations.h

@@ -1,79 +1,89 @@
 // SPDX-License-Identifier: MIT
 #pragma once
 #include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/CompilerDefs.h>
+
+namespace FEXCore::Context {
+class ContextImpl;
+}
 
 namespace FEXCore::CPU {
-enum class RelocationTypes : uint8_t {
+enum class RelocationTypes : uint32_t {
   // 8 byte literal in memory for symbol
   // Aligned to struct RelocNamedSymbolLiteral
   RELOC_NAMED_SYMBOL_LITERAL,
 
   // Fixed size named thunk move
-  // 4 instruction constant generation on AArch64
-  // 64-bit mov on x86-64
+  // 4 instruction constant generation
   // Aligned to struct RelocNamedThunkMove
   RELOC_NAMED_THUNK_MOVE,
 
+  // 8 byte literal (relative to binary base address)
+  RELOC_GUEST_RIP_LITERAL,
+
   // Fixed size guest RIP move
-  // 4 instruction constant generation on AArch64
-  // 64-bit mov on x86-64
-  // Aligned to struct RelocGuestRIPMove
+  // 4 instruction constant generation
+  // Aligned to struct RelocGuestRIP
   RELOC_GUEST_RIP_MOVE,
 };
 
-struct RelocationTypeHeader final {
+struct FEX_PACKED RelocationHeader final {
+  // Offset to the relocated host code data
+  uint64_t Offset {};
+
   RelocationTypes Type;
 };
 
 struct RelocNamedSymbolLiteral final {
-  enum class NamedSymbol : uint8_t {
+  enum class NamedSymbol : uint32_t {
     ///< Thread specific relocations
     // JIT Literal pointers
     SYMBOL_LITERAL_EXITFUNCTION_LINKER,
   };
 
-  RelocationTypeHeader Header {};
+  RelocationHeader Header {};
 
   NamedSymbol Symbol;
 
-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
+  uint32_t Pad[8];
 };
 
 struct RelocNamedThunkMove final {
-  RelocationTypeHeader Header {};
+  RelocationHeader Header {};
 
   // GPR index the constant is being moved to
-  uint8_t RegisterIndex;
+  uint32_t RegisterIndex;
 
   // The thunk SHA256 hash
   IR::SHA256Sum Symbol;
-
-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
 };
 
-struct RelocGuestRIPMove final {
-  RelocationTypeHeader Header {};
+struct RelocGuestRIP final {
+  RelocationHeader Header {};
 
-  // GPR index the constant is being moved to
+  // GPR index the constant is being moved to (for non-literal relocations)
   uint8_t RegisterIndex;
 
-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
+  char Pad[3];
 
-  // The unrelocated RIP that is being moved
+  // The base RIP (to be moved by the register for non-literal relocations).
+  // In a serialized code cache, this is relative to the binary base address.
   uint64_t GuestRIP;
+
+  uint32_t pad2[6] {};
 };
 
 union Relocation {
-  RelocationTypeHeader Header {};
+  RelocationHeader Header {};
 
   RelocNamedSymbolLiteral NamedSymbolLiteral;
   // This makes our union of relocations at least 48 bytes
   // It might be more efficient to not use a union
   RelocNamedThunkMove NamedThunkMove;
 
-  RelocGuestRIPMove GuestRIPMove;
+  RelocGuestRIP GuestRIP;
 };
+
+uint64_t GetNamedSymbolLiteral(FEXCore::Context::ContextImpl&, RelocNamedSymbolLiteral::NamedSymbol);
+
 } // namespace FEXCore::CPU

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

@@ -41,6 +41,7 @@ namespace FEXCore::CPU {
     const auto Op = IROp->C<IR::IROp_##FEXOp>();                                                                                    \
     const auto OpSize = IROp->Size;                                                                                                 \
     const auto Is256Bit = OpSize == IR::OpSize::i256Bit;                                                                            \
+    const auto Is128Bit = OpSize == IR::OpSize::i128Bit;                                                                            \
     LOGMAN_THROW_A_FMT(!Is256Bit || HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__); \
                                                                                                                                     \
     const auto Dst = GetVReg(Node);                                                                                                 \
@@ -49,8 +50,10 @@ namespace FEXCore::CPU {
                                                                                                                                     \
     if (HostSupportsSVE256 && Is256Bit) {                                                                                           \
       ARMOp(Dst.Z(), Vector1.Z(), Vector2.Z());                                                                                     \
-    } else {                                                                                                                        \
+    } else if (Is128Bit) {                                                                                                          \
       ARMOp(Dst.Q(), Vector1.Q(), Vector2.Q());                                                                                     \
+    } else {                                                                                                                        \
+      ARMOp(Dst.D(), Vector1.D(), Vector2.D());                                                                                     \
     }                                                                                                                               \
   }
 
@@ -744,11 +747,11 @@ DEF_OP(VFToIScalarInsert) {
     auto Src = *std::get_if<ARMEmitter::VRegister>(&SrcVar);
 
     switch (RoundMode) {
-    case IR::Round_Nearest: frintn(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Negative_Infinity: frintm(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Positive_Infinity: frintp(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Towards_Zero: frintz(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Host: frinti(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::Nearest: frintn(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::NegInfinity: frintm(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::PosInfinity: frintp(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::TowardsZero: frintz(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::Host: frinti(SubRegSize.Scalar, Dst, Src); break;
     }
   };
 

FEXCore/Source/Interface/Core/LookupCache.cpp

@@ -15,7 +15,7 @@ $end_info$
 
 namespace FEXCore {
 GuestToHostMap::GuestToHostMap()
-  : BlockLinks_mbr {fextl::pmr::get_default_resource()} {
+  : BlockLinks_mbr {"FEXMem_BlockLinks"} {
   BlockLinks_pma = fextl::make_unique<std::pmr::polymorphic_allocator<std::byte>>(&BlockLinks_mbr);
   // Setup our PMR map.
   BlockLinks = BlockLinks_pma->new_object<BlockLinksMapType>();
@@ -24,7 +24,7 @@ GuestToHostMap::GuestToHostMap()
 LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
   : ctx {CTX} {
 
-  TotalCacheSize = ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE + L1_SIZE;
+  TotalCacheSize = ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE + MAX_L1_SIZE;
 
   // Block cache ends up looking like this
   // PageMemoryMap[VirtualMemoryRegion >> 12]
@@ -39,6 +39,10 @@ LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
   // We need one pointer per page of virtual memory
   // At 64GB of virtual memory this will allocate 128MB of virtual memory space
   PagePointer = reinterpret_cast<uintptr_t>(FEXCore::Allocator::VirtualAlloc(TotalCacheSize, false, false));
+  LOGMAN_THROW_A_FMT(PagePointer != -1ULL, "Failed to allocate PagePointer");
+
+  FEXCore::Allocator::VirtualName("FEXMem_Lookup", reinterpret_cast<void*>(PagePointer),
+                                  ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE);
   CTX->SyscallHandler->MarkOvercommitRange(PagePointer, TotalCacheSize);
 
   // Allocate our memory backing our pages
@@ -46,14 +50,21 @@ LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
   // XXX: We can drop down to 16KB if we store 4byte offsets from the code base
   // We currently limit to 128MB of real memory for caching for the total cache size.
   // Can end up being inefficient if we compile a small number of blocks per page
-  PageMemory = PagePointer + ctx->Config.VirtualMemSize / 4096 * 8;
-  LOGMAN_THROW_A_FMT(PageMemory != -1ULL, "Failed to allocate page memory");
+  PageMemory = PagePointer + ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8;
 
   // L1 Cache
   L1Pointer = PageMemory + CODE_SIZE;
-  LOGMAN_THROW_A_FMT(L1Pointer != -1ULL, "Failed to allocate L1Pointer");
+  FEXCore::Allocator::VirtualName("FEXMem_Lookup_L1", reinterpret_cast<void*>(L1Pointer), MAX_L1_SIZE);
 
   VirtualMemSize = ctx->Config.VirtualMemSize;
+
+  if (DynamicL1Cache()) {
+    // Start at minimum size when dynamic.
+    L1PointerMask = MIN_L1_ENTRIES - 1;
+  } else {
+    // Start at maximum instead.
+    L1PointerMask = MAX_L1_ENTRIES - 1;
+  }
 }
 
 LookupCache::~LookupCache() {
@@ -64,31 +75,27 @@ LookupCache::~LookupCache() {
   // These will get freed when their memory allocators are deallocated.
 }
 
-void LookupCache::ClearL2Cache() {
-  auto lk = Shared->AcquireLock();
+void LookupCache::ClearL2Cache(const FEXCore::LookupCacheBaseLockToken& lk) {
   // Clear out the page memory
   // PagePointer and PageMemory are sequential with each other. Clear both at once.
-  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE, false);
+  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer),
+                                      ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE, false);
   AllocateOffset = 0;
 }
 
-void LookupCache::ClearThreadLocalCaches() {
-  auto lk = Shared->AcquireLock();
-
+void LookupCache::ClearThreadLocalCaches(const LookupCacheWriteLockToken&) {
   // Clear L1 and L2 by clearing the full cache.
   FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), TotalCacheSize, false);
+  CachedCodePages.clear();
 }
 
-void LookupCache::ClearCache() {
-  auto lk = Shared->AcquireLock();
-
+void LookupCache::ClearCache(const LookupCacheWriteLockToken& lk) {
   // Clear L1 and L2 by clearing the full cache.
-  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), TotalCacheSize, false);
-
+  ClearThreadLocalCaches(lk);
   Shared->ClearCache(lk);
 }
 
-void GuestToHostMap::ClearCache(const LockToken&) {
+void GuestToHostMap::ClearCache(const LookupCacheWriteLockToken&) {
   // Allocate a new pointer from the BlockLinks pma again.
   BlockLinks = BlockLinks_pma->new_object<BlockLinksMapType>();
   // All code is gone, clear the block list

FEXCore/Source/Interface/Core/LookupCache.h

@@ -2,30 +2,57 @@
 #pragma once
 #include "Interface/Context/Context.h"
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/SHMStats.h>
+#include "Utils/WritePriorityMutex.h"
+
 #include <FEXCore/fextl/map.h>
 #include <FEXCore/fextl/memory_resource.h>
 #include <FEXCore/fextl/robin_map.h>
+#include <FEXCore/fextl/robin_set.h>
 #include <FEXCore/fextl/vector.h>
 #include <FEXCore/fextl/memory_resource.h>
 
 #include <cstdint>
-#include <functional>
 #include <stddef.h>
 #include <utility>
 #include <mutex>
 
 namespace FEXCore {
+struct LookupCacheBaseLockToken {
+protected:
+  // Protected constructor - only derived classes can construct
+  LookupCacheBaseLockToken() = default;
+};
+
+struct LookupCacheWriteLockToken : public LookupCacheBaseLockToken {
+private:
+  // Only constructible by GuestToHostMap
+  friend struct GuestToHostMap;
+  LookupCacheWriteLockToken(FEXCore::Utils::WritePriorityMutex::Mutex& Mutex)
+    : Lock {Mutex} {}
+  std::lock_guard<FEXCore::Utils::WritePriorityMutex::Mutex> Lock;
+};
+
+struct LookupCacheReadLockToken : public LookupCacheBaseLockToken {
+private:
+  // Only constructible by GuestToHostMap
+  friend struct GuestToHostMap;
+  LookupCacheReadLockToken(FEXCore::Utils::WritePriorityMutex::Mutex& Mutex)
+    : Lock {Mutex} {}
+  std::shared_lock<FEXCore::Utils::WritePriorityMutex::Mutex> Lock;
+};
 
 struct GuestToHostMap {
-  std::recursive_mutex WriteLock;
-
-  struct LockToken {
-    std::lock_guard<std::recursive_mutex> Lock;
-  };
+  FEXCore::Utils::WritePriorityMutex::Mutex Lock {};
 
   [[nodiscard]]
-  LockToken AcquireLock() {
-    return LockToken {std::lock_guard {WriteLock}};
+  LookupCacheWriteLockToken AcquireWriteLock() {
+    return LookupCacheWriteLockToken {Lock};
+  }
+
+  [[nodiscard]]
+  LookupCacheReadLockToken AcquireReadLock() {
+    return LookupCacheReadLockToken {Lock};
   }
 
   struct BlockLinkTag {
@@ -49,53 +76,72 @@ struct GuestToHostMap {
   // walking each block member and destructing objects.
   //
   // This makes `BlockLinks` look like a raw pointer that could memory leak, but since it is backed by the MBR, it won't.
-  std::pmr::monotonic_buffer_resource BlockLinks_mbr;
+  fextl::pmr::named_monotonic_page_buffer_resource BlockLinks_mbr;
   using BlockLinksMapType = std::pmr::map<BlockLinkTag, FEXCore::Context::BlockDelinkerFunc>;
   fextl::unique_ptr<std::pmr::polymorphic_allocator<std::byte>> BlockLinks_pma;
   BlockLinksMapType* BlockLinks;
 
-  fextl::robin_map<uint64_t, uint64_t> BlockList;
+  struct BlockEntry {
+    uint64_t HostCode;
+    fextl::vector<uint64_t> CodePages;
+  };
+
+  fextl::robin_map<uint64_t, BlockEntry> BlockList;
 
   fextl::map<uint64_t, fextl::vector<uint64_t>> CodePages;
 
   GuestToHostMap();
 
   // Adds to Guest -> Host code mapping
-  void AddBlockMapping(uint64_t Address, void* HostCode, const LockToken&) {
+  const BlockEntry& AddBlockMapping(uint64_t Address, const fextl::vector<uint64_t>& CodePages, void* HostCode, const LookupCacheWriteLockToken&) {
     // This may replace an existing mapping
     // NOTE: Generally no previous entry should exist, however there is one exception:
     //       If the backend updates the active thread's CodeBuffer, the new associated LookupCache
     //       may already contain the block address. Since is comparatively rare, we'll just leak
     //       one of the two blocks in this case.
-    BlockList[Address] = (uintptr_t)HostCode;
+    return BlockList.insert_or_assign(Address, BlockEntry {(uintptr_t)HostCode, CodePages}).first->second;
   }
 
-  std::optional<uintptr_t> FindBlock(uint64_t Address, const LockToken&) {
+  const BlockEntry* FindBlock(uint64_t Address, const LookupCacheReadLockToken&) {
     auto HostCode = BlockList.find(Address);
     if (HostCode == BlockList.end()) {
-      return std::nullopt;
+      return nullptr;
     }
-    return HostCode->second;
+    return &HostCode->second;
   }
 
-  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address, const LockToken&) {
+  bool Erase(uint64_t Address, const LookupCacheWriteLockToken&) {
     // Sever any links to this block
     auto lower = BlockLinks->lower_bound({Address, nullptr});
     auto upper = BlockLinks->upper_bound({Address, reinterpret_cast<FEXCore::Context::ExitFunctionLinkData*>(UINTPTR_MAX)});
     for (auto it = lower; it != upper; it = BlockLinks->erase(it)) {
-      it->second(Frame, it->first.HostLink);
+      it->second(it->first.HostLink);
     }
 
     // Remove from BlockList
     return BlockList.erase(Address) != 0;
   }
 
+  void InvalidateRange(uint64_t Start, uint64_t Length) {
+    auto lk = AcquireWriteLock();
+
+    auto lower = CodePages.lower_bound(Start >> 12);
+    auto upper = CodePages.upper_bound((Start + Length - 1) >> 12);
+
+    for (auto it = lower; it != upper; it++) {
+      for (const auto& Entry : it->second) {
+        Erase(Entry, lk);
+      }
+    }
+    CodePages.erase(lower, upper);
+  }
+
   void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink,
-                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LockToken&) {
+                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LookupCacheWriteLockToken&) {
     BlockLinks->insert({{GuestDestination, HostLink}, delinker});
   }
 
-  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length, const LockToken&) {
+  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length, const LookupCacheWriteLockToken&) {
     bool rv = false;
 
     for (auto CurrentPage = Start >> 12, EndPage = (Start + Length - 1) >> 12; CurrentPage <= EndPage; CurrentPage++) {
@@ -107,7 +153,7 @@ struct GuestToHostMap {
     return rv;
   }
 
-  void ClearCache(const LockToken&);
+  void ClearCache(const LookupCacheWriteLockToken&);
 };
 
 class LookupCache {
@@ -122,122 +168,199 @@ public:
 
   // Swaps out the underlying GuestToHostMap and clears all associated caches.
   // This interface requires the previous CodeBuffer to be provided despite not using it. This ensures the shared write lock is still valid.
-  void ChangeGuestToHostMapping([[maybe_unused]] CPU::CodeBuffer& Prev, GuestToHostMap& NewMap) {
-    ClearThreadLocalCaches();
+  void ChangeGuestToHostMapping([[maybe_unused]] CPU::CodeBuffer& Prev, GuestToHostMap& NewMap, const LookupCacheWriteLockToken& lk) {
+    ClearThreadLocalCaches(lk);
     Shared = &NewMap;
   }
 
-  uintptr_t FindBlock(uint64_t Address) {
+  uintptr_t FindBlock(FEXCore::Core::InternalThreadState* Thread, uint64_t Address) {
     // Try L1, no lock needed
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
     if (L1Entry.GuestCode == Address) {
       return L1Entry.HostCode;
     }
 
     // L2 and L3 need to be locked
-    auto lk = Shared->AcquireLock();
+    uintptr_t HostPtr {};
+    {
+      std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+        Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheReadLockTime : nullptr);
+      auto lk = Shared->AcquireReadLock();
+      LockTime.reset();
 
-    // Try L2
-    const auto PageIndex = (Address & (VirtualMemSize - 1)) >> 12;
-    const auto PageOffset = Address & (0x0FFF);
+      if (!DisableL2Cache()) {
+        // Try L2
+        const auto PageIndex = (Address & (VirtualMemSize - 1)) >> 12;
+        const auto PageOffset = Address & (0x0FFF);
 
-    const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    auto LocalPagePointer = Pointers[PageIndex];
+        const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+        auto LocalPagePointer = Pointers[PageIndex];
 
-    // Do we a page pointer for this address?
-    if (LocalPagePointer) {
-      // Find there pointer for the address in the blocks
-      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+        // Do we a page pointer for this address?
+        if (LocalPagePointer) {
+          // Find there pointer for the address in the blocks
+          auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
 
-      if (BlockPointers[PageOffset].GuestCode == Address) {
-        L1Entry.GuestCode = Address;
-        L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
-        return L1Entry.HostCode;
+          if (BlockPointers[PageOffset].GuestCode == Address) {
+            L1Entry.GuestCode = Address;
+            L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
+            HostPtr = L1Entry.HostCode;
+          }
+        }
+      }
+
+      if (!HostPtr) {
+        // Try L3
+        auto Entry = Shared->FindBlock(Address, lk);
+        if (Entry) {
+          CacheBlockMapping(Address, *Entry, false, lk);
+          HostPtr = Entry->HostCode;
+        }
       }
     }
 
-    // Try L3
-    auto HostCode = Shared->FindBlock(Address, lk);
-    if (HostCode) {
-      CacheBlockMapping(Address, HostCode.value());
-      return HostCode.value();
+    if (HostPtr && DynamicL1Cache()) {
+      UpdateDynamicL1Stats(Thread);
     }
 
-    // Failed to find
-    return 0;
+    FEXCORE_PROFILE_INSTANT_INCREMENT(Thread, AccumulatedCacheMissCount, 1);
+
+    return HostPtr;
+  }
+
+  void UpdateDynamicL1Stats(FEXCore::Core::InternalThreadState* Thread) {
+    // If host pointer was found in L2 or L3, then add it to the counter.
+    // Keeping track not L1 misses, but specifically L2/L3 hits.
+    ++L2L3CacheHits;
+
+    const auto CurrentTime = std::chrono::system_clock::now();
+    const auto Period = CurrentTime - LastPeriod;
+    if (Period >= SamplePeriod) {
+      // If larger than the sample period then check if we need to increase L1 cache size.
+      const double AveragePerSecond = static_cast<double>(L2L3CacheHits) /
+                                      static_cast<double>(std::chrono::duration_cast<std::chrono::milliseconds>(Period).count()) * 1000.0;
+
+      if (AveragePerSecond >= DynamicL1CacheIncreaseCountHeuristic()) {
+        if (CurrentL1Entries < MAX_L1_ENTRIES) {
+          CurrentL1Entries <<= 1;
+          L1PointerMask = CurrentL1Entries - 1;
+
+          // Update the thread's L1 pointer mask to increase how much cache it uses.
+          // Since we're in C-code, this is safe to update here.
+          Thread->CurrentFrame->State.L1Mask = GetScaledL1PointerMask();
+        }
+      } else if (AveragePerSecond < DynamicL1CacheDecreaseCountHeuristic()) {
+        if (CurrentL1Entries > MIN_L1_ENTRIES) {
+          CurrentL1Entries >>= 1;
+          L1PointerMask = CurrentL1Entries - 1;
+
+          // Madvise the entries that we are dropping. Gives the memory back to the OS.
+          LookupCacheEntry* FirstZeroL1Entry = &reinterpret_cast<LookupCacheEntry*>(L1Pointer)[CurrentL1Entries];
+          size_t ZeroMemorySize = (MAX_L1_ENTRIES - CurrentL1Entries) * sizeof(LookupCacheEntry);
+          FEXCore::Allocator::VirtualDontNeed(FirstZeroL1Entry, ZeroMemorySize, false);
+
+          // Update the thread's L1 pointer mask to increase how much cache it uses.
+          // Since we're in C-code, this is safe to update here.
+          Thread->CurrentFrame->State.L1Mask = GetScaledL1PointerMask();
+        }
+      }
+
+      // Update Last period to start again.
+      LastPeriod = CurrentTime;
+      L2L3CacheHits = 0;
+    }
   }
 
   GuestToHostMap* Shared = nullptr;
 
   // Appends a list of Block {Address} to CodePages [Start, Start + Length)
   // Returns true if new pages are marked as containing code
-  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length) {
-    auto lk = Shared->AcquireLock();
+  bool AddBlockExecutableRange(FEXCore::Core::InternalThreadState* Thread, const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length) {
+    std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+      Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheWriteLockTime : nullptr);
+    auto lk = Shared->AcquireWriteLock();
+    LockTime.reset();
+
     return Shared->AddBlockExecutableRange(Addresses, Start, Length, lk);
   }
 
   // Adds to Guest -> Host code mapping
-  void AddBlockMapping(uint64_t Address, void* HostCode) {
-    auto lk = Shared->AcquireLock();
+  void AddBlockMapping(FEXCore::Core::InternalThreadState* Thread, uint64_t Address, const fextl::vector<uint64_t>& CodePages, void* HostCode) {
+    std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+      Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheWriteLockTime : nullptr);
+    auto lk = Shared->AcquireWriteLock();
+    LockTime.reset();
 
-    Shared->AddBlockMapping(Address, HostCode, lk);
+    const auto& Entry = Shared->AddBlockMapping(Address, CodePages, HostCode, lk);
 
     // There is no need to update L1 or L2, they will get updated on first lookup
     // However, adding to L1 here increases performance
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
-    L1Entry.GuestCode = Address;
-    L1Entry.HostCode = (uintptr_t)HostCode;
+    CacheBlockMapping(Address, Entry, true, lk);
   }
 
-  // NOTE: It's the caller's responsibility to call Erase() for all other
-  //       GuestToHostMaps that share the same LookupCache. Otherwise, the
-  //       L1/L2 caches will contain stale references to deallocated memory.
-  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address) {
-    auto lk = Shared->AcquireLock();
-
-    bool ErasedAny = Shared->Erase(Frame, Address, lk);
-
+  // Invalidates L1/L2 for a given guest block
+  void InvalidateCache(uint64_t Address, const LookupCacheWriteLockToken& lk) {
     // Do L1
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
     if (L1Entry.GuestCode == Address) {
       L1Entry.GuestCode = 0;
-      ErasedAny = true;
       // Leave L1Entry.HostCode as is, so that concurrent lookups won't read a null pointer
       // This is a soft guarantee for cross thread invalidation, as atomics are not used
       // and it hasn't been thoroughly tested
     }
 
-    // Do full map
-    Address = Address & (VirtualMemSize - 1);
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
+    if (!DisableL2Cache()) {
+      // Do full map
+      Address = Address & (VirtualMemSize - 1);
+      uint64_t PageOffset = Address & (0x0FFF);
+      Address >>= 12;
 
-    uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // Page for this code didn't even exist, nothing to do
-      return ErasedAny;
+      uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+      uint64_t LocalPagePointer = Pointers[Address];
+      if (!LocalPagePointer) {
+        // Page for this code didn't even exist, nothing to do
+        return;
+      }
+
+      // Page exists, just set the offset to zero
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+      BlockPointers[PageOffset].GuestCode = 0;
+      BlockPointers[PageOffset].HostCode = 0;
     }
-
-    // Page exists, just set the offset to zero
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
-    BlockPointers[PageOffset].GuestCode = 0;
-    BlockPointers[PageOffset].HostCode = 0;
-    return true;
   }
 
-  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink, const FEXCore::Context::BlockDelinkerFunc& delinker) {
-    auto lk = Shared->AcquireLock();
+  // Invalidates all L1/L2 entries for all guest block that intersect the given range
+  bool InvalidateCacheRange(uint64_t Start, uint64_t Length) {
+    auto lk = Shared->AcquireWriteLock();
+
+    auto lower = CachedCodePages.lower_bound(Start >> 12);
+    auto upper = CachedCodePages.upper_bound((Start + Length - 1) >> 12);
+
+    for (auto it = lower; it != upper; it++) {
+      for (const auto& Entry : it->second) {
+        InvalidateCache(Entry, lk);
+      }
+    }
+    bool ret = upper != lower;
+    CachedCodePages.erase(lower, upper);
+    return ret;
+  }
+
+  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink,
+                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LookupCacheWriteLockToken& lk) {
     Shared->AddBlockLink(GuestDestination, HostLink, delinker, lk);
   }
 
-  void ClearCache();
-  void ClearL2Cache();
-  void ClearThreadLocalCaches();
+  void ClearCache(const LookupCacheWriteLockToken&);
+  void ClearL2Cache(const LookupCacheBaseLockToken&);
+  void ClearThreadLocalCaches(const LookupCacheWriteLockToken&);
 
   uintptr_t GetL1Pointer() const {
     return L1Pointer;
   }
+  uintptr_t GetScaledL1PointerMask() const {
+    return L1PointerMask << FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry));
+  }
   uintptr_t GetPagePointer() const {
     return PagePointer;
   }
@@ -245,9 +368,6 @@ public:
     return VirtualMemSize;
   }
 
-  constexpr static size_t L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
-  constexpr static size_t L1_ENTRIES_MASK = L1_ENTRIES - 1;
-
   // This needs to be taken before reads or writes to L2, L3, CodePages,
   // and before writes to L1. Concurrent access from a thread that this LookupCache doesn't belong to
   // may only happen during cross thread invalidation (::Erase).
@@ -255,45 +375,52 @@ public:
   // Some care is taken so that L1 lookups can be done without locks, and even tearing is unlikely to lead to a crash.
   // This approach has not been fully vetted yet.
   // Also note that L1 lookups might be inlined in the JIT Dispatcher and/or block ends.
-  auto AcquireLock() {
-    return Shared->AcquireLock();
+  auto AcquireWriteLock() {
+    return Shared->AcquireWriteLock();
   }
 
 private:
-  void CacheBlockMapping(uint64_t Address, uintptr_t HostCode) {
-    // Do L1
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
-    L1Entry.GuestCode = Address;
-    L1Entry.HostCode = HostCode;
-
-    // Do ful map
-    auto FullAddress = Address;
-    Address = Address & (VirtualMemSize - 1);
-
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
-    uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // We don't have a page pointer for this address
-      // Allocate one now if we can
-      uintptr_t NewPageBacking = AllocateBackingForPage();
-      if (!NewPageBacking) {
-        // Couldn't allocate, clear L2 and retry
-        ClearL2Cache();
-        CacheBlockMapping(Address, HostCode);
-        return;
-      }
-      Pointers[Address] = NewPageBacking;
-      LocalPagePointer = NewPageBacking;
+  void CacheBlockMapping(uint64_t Address, const GuestToHostMap::BlockEntry& Entry, bool L1Only, const LookupCacheBaseLockToken& lk) {
+    for (const auto& CodePage : Entry.CodePages) {
+      CachedCodePages[CodePage >> 12].insert(Address);
     }
 
-    // Add the new pointer to the page block
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+    // Do L1
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
+    L1Entry.GuestCode = Address;
+    L1Entry.HostCode = Entry.HostCode;
 
-    // This silently replaces existing mappings
-    BlockPointers[PageOffset].GuestCode = FullAddress;
-    BlockPointers[PageOffset].HostCode = HostCode;
+    if (!DisableL2Cache() && !L1Only) {
+      // Do ful map
+      auto FullAddress = Address;
+      Address = Address & (VirtualMemSize - 1);
+
+      uint64_t PageOffset = Address & (0x0FFF);
+      Address >>= 12;
+
+      uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+      uint64_t LocalPagePointer = Pointers[Address];
+      if (!LocalPagePointer) {
+        // We don't have a page pointer for this address
+        // Allocate one now if we can
+        uintptr_t NewPageBacking = AllocateBackingForPage();
+        if (!NewPageBacking) {
+          // Couldn't allocate, clear L2 and retry
+          ClearL2Cache(lk);
+          CacheBlockMapping(Address, Entry, false, lk);
+          return;
+        }
+        Pointers[Address] = NewPageBacking;
+        LocalPagePointer = NewPageBacking;
+      }
+
+      // Add the new pointer to the page block
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+
+      // This silently replaces existing mappings
+      BlockPointers[PageOffset].GuestCode = FullAddress;
+      BlockPointers[PageOffset].HostCode = Entry.HostCode;
+    }
   }
 
   uintptr_t AllocateBackingForPage() {
@@ -310,19 +437,38 @@ private:
     return PageMemory + NewBase;
   }
 
+  // Maps from a page index to all blocks in the page that have at some point been fetched into L1/L2
+  fextl::map<uint64_t, fextl::robin_set<uint64_t>> CachedCodePages;
+
   uintptr_t PagePointer;
   uintptr_t PageMemory;
   uintptr_t L1Pointer;
+  uintptr_t L1PointerMask;
 
   size_t TotalCacheSize;
 
+  // Start with 8k entries in L1 to give 128KB of L1 cache to each thread.
+  // Max out at 1 million entries to give each thread 16MB of L1 cache maximum.
+  constexpr static size_t MIN_L1_ENTRIES = 8 * 1024;        // Must be a power of 2
+  constexpr static size_t MAX_L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
+
   constexpr static size_t CODE_SIZE = 128 * 1024 * 1024;
-  constexpr static size_t SIZE_PER_PAGE = 4096 * sizeof(LookupCacheEntry);
-  constexpr static size_t L1_SIZE = L1_ENTRIES * sizeof(LookupCacheEntry);
+  constexpr static size_t SIZE_PER_PAGE = FEXCore::Utils::FEX_PAGE_SIZE * sizeof(LookupCacheEntry);
+  constexpr static size_t MAX_L1_SIZE = MAX_L1_ENTRIES * sizeof(LookupCacheEntry);
 
   size_t AllocateOffset {};
 
   FEXCore::Context::ContextImpl* ctx;
   uint64_t VirtualMemSize {};
+
+  size_t CurrentL1Entries = MIN_L1_ENTRIES;
+  uint64_t L2L3CacheHits {};
+  std::chrono::time_point<std::chrono::system_clock> LastPeriod {};
+  constexpr static std::chrono::seconds SamplePeriod {1};
+  FEX_CONFIG_OPT(DynamicL1CacheIncreaseCountHeuristic, DYNAMICL1CACHEINCREASECOUNTHEURISTIC);
+  FEX_CONFIG_OPT(DynamicL1CacheDecreaseCountHeuristic, DYNAMICL1CACHEDECREASECOUNTHEURISTIC);
+
+  FEX_CONFIG_OPT(DynamicL1Cache, DYNAMICL1CACHE);
+  FEX_CONFIG_OPT(DisableL2Cache, DISABLEL2CACHE);
 };
 } // namespace FEXCore

FEXCore/Source/Interface/Core/OpcodeDispatcher.h

@@ -139,27 +139,28 @@ public:
     FlushRegisterCache();
     return _Jump(_TargetBlock);
   }
-  IRPair<IROp_CondJump> CondJump(Ref _Cmp1, Ref _Cmp2, Ref _TrueBlock, Ref _FalseBlock, CondClassType _Cond = {COND_NEQ},
+  IRPair<IROp_CondJump> CondJump(Ref _Cmp1, Ref _Cmp2, Ref _TrueBlock, Ref _FalseBlock, CondClass _Cond = CondClass::NEQ,
                                  IR::OpSize _CompareSize = OpSize::iInvalid) {
     FlushRegisterCache();
     return _CondJump(_Cmp1, _Cmp2, _TrueBlock, _FalseBlock, _Cond, _CompareSize);
   }
-  IRPair<IROp_CondJump> CondJump(Ref ssa0, CondClassType cond = {COND_NEQ}) {
+  IRPair<IROp_CondJump> CondJump(Ref ssa0, CondClass cond = CondClass::NEQ) {
     FlushRegisterCache();
     return _CondJump(ssa0, cond);
   }
-  IRPair<IROp_CondJump> CondJump(Ref ssa0, Ref ssa1, Ref ssa2, CondClassType cond = {COND_NEQ}) {
+  IRPair<IROp_CondJump> CondJump(Ref ssa0, Ref ssa1, Ref ssa2, CondClass cond = CondClass::NEQ) {
     FlushRegisterCache();
     return _CondJump(ssa0, ssa1, ssa2, cond);
   }
-  IRPair<IROp_CondJump> CondJumpNZCV(CondClassType Cond) {
+  IRPair<IROp_CondJump> CondJumpNZCV(CondClass Cond) {
     FlushRegisterCache();
     return _CondJump(InvalidNode, InvalidNode, InvalidNode, InvalidNode, Cond, OpSize::iInvalid, true);
   }
   IRPair<IROp_CondJump> CondJumpBit(Ref Src, unsigned Bit, bool Set) {
     FlushRegisterCache();
     auto InlineConst = _InlineConstant(Bit);
-    return _CondJump(Src, InlineConst, InvalidNode, InvalidNode, {Set ? COND_TSTNZ : COND_TSTZ}, OpSize::iInvalid, false);
+    auto Cond = Set ? CondClass::TSTNZ : CondClass::TSTZ;
+    return _CondJump(Src, InlineConst, InvalidNode, InvalidNode, Cond, OpSize::iInvalid, false);
   }
   IRPair<IROp_ExitFunction> ExitFunction(Ref NewRIP, BranchHint Hint = BranchHint::None) {
     FlushRegisterCache();
@@ -251,7 +252,7 @@ public:
     auto ExitBlock = CreateNewCodeBlockAfter(BackwardBlock);
 
     auto DF = GetRFLAG(X86State::RFLAG_DF_RAW_LOC);
-    CondJump(DF, Zero, ForwardBlock, BackwardBlock, {COND_EQ});
+    CondJump(DF, Zero, ForwardBlock, BackwardBlock, CondClass::EQ);
 
     for (auto D = 0; D < 2; ++D) {
       SetCurrentCodeBlock(D ? BackwardBlock : ForwardBlock);
@@ -564,7 +565,7 @@ public:
   template<IR::OpSize DstElementSize, IR::OpSize SrcElementSize>
   void AVXInsertScalar_CVT_Float_To_Float(OpcodeArgs);
 
-  RoundType TranslateRoundType(uint8_t Mode);
+  RoundMode TranslateRoundType(uint8_t Mode);
 
   template<IR::OpSize ElementSize>
   void InsertScalarRound(OpcodeArgs);
@@ -758,7 +759,6 @@ public:
   void X87FXTRACT(OpcodeArgs);
   void X87FYL2X(OpcodeArgs, bool IsFYL2XP1);
   void X87LDENV(OpcodeArgs);
-  void X87LDSW(OpcodeArgs);
   void X87ModifySTP(OpcodeArgs, bool Inc);
   void X87OpHelper(OpcodeArgs, FEXCore::IR::IROps IROp, bool ZeroC2);
 
@@ -907,6 +907,10 @@ public:
   void VPCLMULQDQOp(OpcodeArgs);
 
   void CRC32(OpcodeArgs);
+  void Extrq_imm(OpcodeArgs);
+  void Insertq_imm(OpcodeArgs);
+  void Extrq(OpcodeArgs);
+  void Insertq(OpcodeArgs);
 
   void BreakOp(OpcodeArgs, FEXCore::IR::BreakDefinition BreakDefinition);
   void UnimplementedOp(OpcodeArgs);
@@ -985,7 +989,6 @@ public:
   void AVX128_VPSIGN(OpcodeArgs, IR::OpSize ElementSize);
   void AVX128_UCOMISx(OpcodeArgs, IR::OpSize ElementSize);
   void AVX128_VectorScalarInsertALU(OpcodeArgs, FEXCore::IR::IROps IROp, IR::OpSize ElementSize);
-  Ref AVX128_VFCMPImpl(IR::OpSize ElementSize, Ref Src1, Ref Src2, uint8_t CompType);
   void AVX128_VFCMP(OpcodeArgs, IR::OpSize ElementSize);
   void AVX128_InsertScalarFCMP(OpcodeArgs, IR::OpSize ElementSize);
   void AVX128_MOVBetweenGPR_FPR(OpcodeArgs);
@@ -1004,9 +1007,7 @@ public:
   void AVX128_VINSERT(OpcodeArgs);
   void AVX128_VINSERTPS(OpcodeArgs);
 
-  Ref AVX128_PHSUBImpl(Ref Src1, Ref Src2, size_t ElementSize);
   void AVX128_VPHSUB(OpcodeArgs, IR::OpSize ElementSize);
-
   void AVX128_VPHSUBSW(OpcodeArgs);
 
   void AVX128_VADDSUBP(OpcodeArgs, IR::OpSize ElementSize);
@@ -1098,8 +1099,8 @@ public:
   void AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx);
   void AVX128_VFMAddSubImpl(OpcodeArgs, bool AddSub, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx);
 
-  RefPair AVX128_VPGatherQPSImpl(Ref Dest, Ref Mask, RefVSIB VSIB);
-  RefPair AVX128_VPGatherImpl(OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest, RefPair Mask, RefVSIB VSIB);
+  RefPair AVX128_VPGatherQPSImpl(OpcodeArgs, Ref Dest, Ref Mask, RefVSIB VSIB);
+  RefPair AVX128_VPGatherImpl(OpcodeArgs, OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest, RefPair Mask, RefVSIB VSIB);
 
   void AVX128_VPGATHER(OpcodeArgs, OpSize AddrElementSize);
 
@@ -1109,8 +1110,8 @@ public:
   // End of AVX 128-bit implementation
 
   // AVX 256-bit operations
-  void StoreResult_WithAVXInsert(VectorOpType Type, FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, Ref Value,
-                                 IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+  void StoreResult_WithAVXInsert(VectorOpType Type, RegClass Class, FEXCore::X86Tables::DecodedOp Op, Ref Value,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
     if (Op->Dest.IsGPR() && Op->Dest.Data.GPR.GPR >= X86State::REG_XMM_0 && Op->Dest.Data.GPR.GPR <= X86State::REG_XMM_15 &&
         GetGuestVectorLength() == OpSize::i256Bit && Type == VectorOpType::SSE) {
       const auto gpr = Op->Dest.Data.GPR.GPR;
@@ -1161,7 +1162,7 @@ public:
     }
   }
 
-  void StoreContextHelper(IR::OpSize Size, RegisterClassType Class, Ref Value, uint32_t Offset) {
+  void StoreContextHelper(IR::OpSize Size, RegClass Class, Ref Value, uint32_t Offset) {
     // For i128Bit, we won't see a normal Constant to inline, but as a special
     // case we can replace with a 2x64-bit store which can use inline zeroes.
     if (Size == OpSize::i128Bit) {
@@ -1173,7 +1174,7 @@ public:
 
         if (Const->Constant == IR::NamedVectorConstant::NAMED_VECTOR_ZERO) {
           Ref Zero = _Constant(0);
-          Ref STP = _StoreContextPair(IR::OpSize::i64Bit, GPRClass, Zero, Zero, Offset);
+          Ref STP = _StoreContextPair(IR::OpSize::i64Bit, RegClass::GPR, Zero, Zero, Offset);
 
           // XXX: This works around InlineConstant not having an associated
           // register class, else we'd just do InlineConstant above.
@@ -1229,16 +1230,16 @@ public:
 
       if (Index >= GPR0Index && Index <= GPR15Index) {
         Ref R = _StoreRegister(Value, GPRSize);
-        R->Reg = PhysicalRegister(GPRFixedClass, Index - GPR0Index).Raw;
+        R->Reg = PhysicalRegister(RegClass::GPRFixed, Index - GPR0Index).Raw;
       } else if (Index == PFIndex) {
         _StorePF(Value, GPRSize);
       } else if (Index == AFIndex) {
         _StoreAF(Value, GPRSize);
       } else if (Index >= FPR0Index && Index <= FPR15Index) {
         Ref R = _StoreRegister(Value, VectorSize);
-        R->Reg = PhysicalRegister(FPRFixedClass, Index - FPR0Index).Raw;
+        R->Reg = PhysicalRegister(RegClass::FPRFixed, Index - FPR0Index).Raw;
       } else if (Index == DFIndex) {
-        _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(Core::CPUState, flags[X86State::RFLAG_DF_RAW_LOC]));
+        _StoreContextGPR(OpSize::i8Bit, Value, offsetof(Core::CPUState, flags[X86State::RFLAG_DF_RAW_LOC]));
       } else {
         bool Partial = RegCache.Partial & (1ull << Index);
         auto Size = Partial ? OpSize::i64Bit : CacheIndexToOpSize(Index);
@@ -1263,7 +1264,7 @@ public:
           StoreContextHelper(Size, Class, Value, Offset);
           // If Partial and MMX register, then we need to store all 1s in bits 64-80
           if (Partial && Index >= MM0Index && Index <= MM7Index) {
-            _StoreContext(OpSize::i16Bit, IR::GPRClass, Constant(0xFFFF), Offset + 8);
+            _StoreContextGPR(OpSize::i16Bit, Constant(0xFFFF), Offset + 8);
           }
         }
       }
@@ -1553,23 +1554,63 @@ private:
 
   AddressMode DecodeAddress(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, MemoryAccessType AccessType, bool IsLoad);
 
-  Ref LoadSource(RegisterClassType Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+  Ref LoadSource(RegClass Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
                  const LoadSourceOptions& Options = {});
-  Ref LoadSource_WithOpSize(RegisterClassType Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand,
-                            IR::OpSize OpSize, uint32_t Flags, const LoadSourceOptions& Options = {});
-  void StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op,
-                              const FEXCore::X86Tables::DecodedOperand& Operand, const Ref Src, IR::OpSize OpSize, IR::OpSize Align,
-                              MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
-  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, const FEXCore::X86Tables::DecodedOperand& Operand,
-                   const Ref Src, IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
-  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, const Ref Src, IR::OpSize Align,
+  Ref LoadSourceGPR(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+                    const LoadSourceOptions& Options = {}) {
+    return LoadSource(RegClass::GPR, Op, Operand, Flags, Options);
+  }
+  Ref LoadSourceFPR(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+                    const LoadSourceOptions& Options = {}) {
+    return LoadSource(RegClass::FPR, Op, Operand, Flags, Options);
+  }
+
+  Ref LoadSource_WithOpSize(RegClass Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize,
+                            uint32_t Flags, const LoadSourceOptions& Options = {});
+  Ref LoadSourceGPR_WithOpSize(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize, uint32_t Flags,
+                               const LoadSourceOptions& Options = {}) {
+    return LoadSource_WithOpSize(RegClass::GPR, Op, Operand, OpSize, Flags, Options);
+  }
+  Ref LoadSourceFPR_WithOpSize(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize, uint32_t Flags,
+                               const LoadSourceOptions& Options = {}) {
+    return LoadSource_WithOpSize(RegClass::FPR, Op, Operand, OpSize, Flags, Options);
+  }
+
+  void StoreResult_WithOpSize(RegClass Class, X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                              IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR_WithOpSize(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult_WithOpSize(RegClass::GPR, Op, Operand, Src, OpSize, Align, AccessType);
+  }
+  void StoreResultFPR_WithOpSize(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult_WithOpSize(RegClass::FPR, Op, Operand, Src, OpSize, Align, AccessType);
+  }
+
+  void StoreResult(RegClass Class, X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align,
                    MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align = OpSize::iInvalid,
+                      MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::GPR, Op, Operand, Src, Align, AccessType);
+  }
+  void StoreResultFPR(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align = OpSize::iInvalid,
+                      MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::FPR, Op, Operand, Src, Align, AccessType);
+  }
+
+  void StoreResult(RegClass Class, X86Tables::DecodedOp Op, Ref Src, OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR(X86Tables::DecodedOp Op, Ref Src, OpSize Align = OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::GPR, Op, Src, Align, AccessType);
+  }
+  void StoreResultFPR(X86Tables::DecodedOp Op, Ref Src, OpSize Align = OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::FPR, Op, Src, Align, AccessType);
+  }
 
   // In several instances, it's desirable to get a base address with the segment offset
   // applied to it. This pulls all the common-case appending into a single set of functions.
   [[nodiscard]]
   Ref MakeSegmentAddress(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize) {
-    Ref Mem = LoadSource_WithOpSize(GPRClass, Op, Operand, OpSize, Op->Flags, {.LoadData = false});
+    Ref Mem = LoadSourceGPR_WithOpSize(Op, Operand, OpSize, Op->Flags, {.LoadData = false});
     return AppendSegmentOffset(Mem, Op->Flags);
   }
   [[nodiscard]]
@@ -1614,6 +1655,9 @@ private:
     return IR::SizeToOpSize(GetSrcSize(Op));
   }
 
+  [[nodiscard]]
+  IR::OpSize GetStringOpSize(X86Tables::DecodedOp Op) const;
+
   // Set flag tracking to prepare for an operation that directly writes NZCV.
   void HandleNZCVWrite() {
     CachedNZCV = nullptr;
@@ -1805,14 +1849,15 @@ private:
       // For DF, we need to transform 0/1 into 1/-1
       StoreDF(_SubShift(OpSize::i64Bit, Constant(1), Value, ShiftType::LSL, 1));
     } else if (BitOffset == FEXCore::X86State::RFLAG_TF_RAW_LOC) {
-      auto PackedTF = _LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      auto PackedTF = _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
       // An exception should still be raised after an instruction that unsets TF, leave the unblocked bit set but unset
       // the TF bit to cause such behaviour. The handling code at the start of the next block will then unset the
       // unblocked bit before raising the exception.
-      auto NewPackedTF = _Select(FEXCore::IR::COND_EQ, Value, Constant(0), _And(OpSize::i32Bit, PackedTF, Constant(~1)), Constant(1));
-      _StoreContext(OpSize::i8Bit, GPRClass, NewPackedTF, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      auto NewPackedTF =
+        _Select(OpSize::i64Bit, OpSize::i64Bit, CondClass::EQ, Value, Constant(0), _And(OpSize::i32Bit, PackedTF, Constant(~1)), Constant(1));
+      _StoreContextGPR(OpSize::i8Bit, NewPackedTF, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
     } else {
-      _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      _StoreContextGPR(OpSize::i8Bit, Value, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
     }
   }
 
@@ -1838,12 +1883,12 @@ private:
   }
 
   [[nodiscard]]
-  static CondClassType CondForNZCVBit(unsigned BitOffset, bool Invert) {
+  static CondClass CondForNZCVBit(unsigned BitOffset, bool Invert) {
     switch (BitOffset) {
-    case X86State::RFLAG_SF_RAW_LOC: return {Invert ? COND_PL : COND_MI};
-    case X86State::RFLAG_ZF_RAW_LOC: return {Invert ? COND_NEQ : COND_EQ};
-    case X86State::RFLAG_CF_RAW_LOC: return {Invert ? COND_ULT : COND_UGE};
-    case X86State::RFLAG_OF_RAW_LOC: return {Invert ? COND_FNU : COND_FU};
+    case X86State::RFLAG_SF_RAW_LOC: return Invert ? CondClass::PL : CondClass::MI;
+    case X86State::RFLAG_ZF_RAW_LOC: return Invert ? CondClass::NEQ : CondClass::EQ;
+    case X86State::RFLAG_CF_RAW_LOC: return Invert ? CondClass::ULT : CondClass::UGE;
+    case X86State::RFLAG_OF_RAW_LOC: return Invert ? CondClass::FNU : CondClass::FU;
     default: FEX_UNREACHABLE;
     }
   }
@@ -1874,11 +1919,11 @@ private:
   }
 
   [[nodiscard]]
-  static RegisterClassType CacheIndexClass(int Index) {
+  static RegClass CacheIndexClass(int Index) {
     if ((Index >= MM0Index && Index <= MM7Index) || Index >= FPR0Index) {
-      return FPRClass;
+      return RegClass::FPR;
     } else {
-      return GPRClass;
+      return RegClass::GPR;
     }
   }
 
@@ -1910,14 +1955,14 @@ private:
     RegCache.Written &= ~Bit;
   }
 
-  Ref LoadRegCache(uint64_t Offset, uint8_t Index, RegisterClassType RegClass, IR::OpSize Size) {
+  Ref LoadRegCache(uint64_t Offset, uint8_t Index, RegClass Class, IR::OpSize Size) {
     LOGMAN_THROW_A_FMT(Index < 64, "valid index");
     uint64_t Bit = (1ull << (uint64_t)Index);
 
     if (Size == OpSize::i128Bit && (RegCache.Partial & Bit)) {
       // We need to load the full register extend if we previously did a partial access.
       Ref Value = RegCache.Value[Index];
-      Ref Full = _LoadContext(Size, RegClass, Offset);
+      Ref Full = _LoadContext(Size, Class, Offset);
 
       // If we did a partial store, we're inserting into the full register
       if (RegCache.Written & Bit) {
@@ -1931,7 +1976,7 @@ private:
       if (Index == DFIndex) {
         RegCache.Value[Index] = _LoadDF();
       } else if ((Index >= MM0Index && Index <= MM7Index) || Index >= AVXHigh0Index) {
-        RegCache.Value[Index] = _LoadContext(Size, RegClass, Offset);
+        RegCache.Value[Index] = _LoadContext(Size, Class, Offset);
 
         // We may have done a partial load, this requires special handling.
         if (Size == OpSize::i64Bit) {
@@ -1942,7 +1987,7 @@ private:
       } else if (Index == AFIndex) {
         RegCache.Value[Index] = _LoadAF(Size);
       } else {
-        RegCache.Value[Index] = _LoadRegister(Offset, RegClass, Size);
+        RegCache.Value[Index] = _LoadRegister(Offset, Class, Size);
       }
 
       RegCache.Cached |= Bit;
@@ -1951,21 +1996,21 @@ private:
     return RegCache.Value[Index];
   }
 
-  RefPair AllocatePair(FEXCore::IR::RegisterClassType Class, IR::OpSize Size) {
-    if (Class == FPRClass) {
+  RefPair AllocatePair(RegClass Class, IR::OpSize Size) {
+    if (Class == RegClass::FPR) {
       return {_AllocateFPR(Size, Size), _AllocateFPR(Size, Size)};
     } else {
       return {_AllocateGPR(false), _AllocateGPR(false)};
     }
   }
 
-  RefPair LoadContextPair_Uncached(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, unsigned Offset) {
+  RefPair LoadContextPair_Uncached(RegClass Class, IR::OpSize Size, unsigned Offset) {
     RefPair Values = AllocatePair(Class, Size);
     _LoadContextPair(Size, Class, Offset, Values.Low, Values.High);
     return Values;
   }
 
-  RefPair LoadRegCachePair(uint64_t Offset, uint8_t Index, RegisterClassType RegClass, IR::OpSize Size) {
+  RefPair LoadRegCachePair(uint64_t Offset, uint8_t Index, RegClass Class, IR::OpSize Size) {
     LOGMAN_THROW_A_FMT(Index != DFIndex, "must be pairable");
     LOGMAN_THROW_A_FMT(Size != IR::OpSize::iUnsized, "Invalid size!");
 
@@ -1973,7 +2018,7 @@ private:
     uint64_t Bits = (3ull << (uint64_t)Index);
     const auto SizeInt = IR::OpSizeToSize(Size);
     if (((RegCache.Partial | RegCache.Cached) & Bits) == 0 && ((Offset / SizeInt) < 64)) {
-      auto Values = LoadContextPair_Uncached(RegClass, Size, Offset);
+      auto Values = LoadContextPair_Uncached(Class, Size, Offset);
       RegCache.Value[Index] = Values.Low;
       RegCache.Value[Index + 1] = Values.High;
       RegCache.Cached |= Bits;
@@ -1985,13 +2030,13 @@ private:
 
     // Fallback on a pair of loads
     return {
-      .Low = LoadRegCache(Offset, Index, RegClass, Size),
-      .High = LoadRegCache(Offset + SizeInt, Index + 1, RegClass, Size),
+      .Low = LoadRegCache(Offset, Index, Class, Size),
+      .High = LoadRegCache(Offset + SizeInt, Index + 1, Class, Size),
     };
   }
 
   Ref LoadGPR(uint8_t Reg) {
-    return LoadRegCache(Reg, GPR0Index + Reg, GPRClass, GetGPROpSize());
+    return LoadRegCache(Reg, GPR0Index + Reg, RegClass::GPR, GetGPROpSize());
   }
 
   Ref LoadContext(IR::OpSize Size, uint8_t Index) {
@@ -2007,7 +2052,7 @@ private:
   }
 
   Ref LoadXMMRegister(uint8_t Reg) {
-    return LoadRegCache(Reg, FPR0Index + Reg, FPRClass, GetGuestVectorLength());
+    return LoadRegCache(Reg, FPR0Index + Reg, RegClass::FPR, GetGuestVectorLength());
   }
 
   Ref LoadDF() {
@@ -2062,7 +2107,7 @@ private:
       // Recover the sign bit, it is the logical DF value
       return _Lshr(OpSize::i64Bit, LoadDF(), Constant(63));
     } else {
-      return _LoadContext(OpSize::i8Bit, GPRClass, offsetof(Core::CPUState, flags[BitOffset]));
+      return _LoadContextGPR(OpSize::i8Bit, offsetof(Core::CPUState, flags[BitOffset]));
     }
   }
 
@@ -2079,18 +2124,18 @@ private:
   }
 
   // Safe version of NZCVSelect that handles inverted carries automatically.
-  Ref NZCVSelect(OpSize OpSize, CondClassType Cond, Ref TrueV, Ref FalseV, bool CarryIsInverted = false) {
+  Ref NZCVSelect(OpSize OpSize, CondClass Cond, Ref TrueV, Ref FalseV, bool CarryIsInverted = false) {
     switch (Cond) {
-    case IR::COND_UGE: /* cs */
-    case IR::COND_ULT: /* cc */
+    case CondClass::UGE: /* cs */
+    case CondClass::ULT: /* cc */
       // Invert the condition to match our expectations.
       if (CarryIsInverted != CFInverted) {
-        Cond = {Cond == COND_UGE ? COND_ULT : COND_UGE};
+        Cond = (Cond == CondClass::UGE) ? CondClass::ULT : CondClass::UGE;
       }
       break;
 
-    case IR::COND_UGT: /* hi */
-    case IR::COND_ULE: /* ls */
+    case CondClass::UGT: /* hi */
+    case CondClass::ULE: /* ls */
       // No clever optimization we can do here, rectify carry itself.
       RectifyCarryInvert(CarryIsInverted);
       break;
@@ -2179,7 +2224,7 @@ private:
 
     HandleNZCV_RMW();
     CalculatePF(_ShiftFlags(OpSizeFromSrc(Op), Result, Dest, Shift, Src, OldPF, CFInverted));
-    StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+    StoreResultGPR(Op, Result);
   }
 
   // Helper to derive Dest by a given builder-using Expression with the opcode
@@ -2252,8 +2297,7 @@ private:
     CachedIndexedNamedVectorConstants.clear();
   }
 
-  std::optional<CondClassType> DecodeNZCVCondition(uint8_t OP);
-  Ref SelectBit(Ref Cmp, IR::OpSize ResultSize, Ref TrueValue, Ref FalseValue);
+  std::optional<CondClass> DecodeNZCVCondition(uint8_t OP);
   Ref SelectCC0All1(uint8_t OP);
 
   /**
@@ -2267,8 +2311,8 @@ private:
     if (Size != OpSize::i32Bit) {
       return;
     }
-    auto Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags);
-    StoreResult(GPRClass, Op, Dest, OpSize::iInvalid);
+    auto Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags);
+    StoreResultGPR(Op, Dest);
   }
 
   using ZeroShiftFunctionPtr = void (OpDispatchBuilder::*)(FEXCore::X86Tables::DecodedOp Op);
@@ -2303,7 +2347,7 @@ private:
 
     ///< Jump to zeroshift block or end block depending on if it was provided.
     IRPair<IROp_CodeBlock> TailHandling = ZeroShiftResult ? ZeroShiftBlock : EndBlock;
-    CondJump(Shift, Zero, TailHandling, SetBlock, {COND_EQ});
+    CondJump(Shift, Zero, TailHandling, SetBlock, CondClass::EQ);
 
     SetCurrentCodeBlock(SetBlock);
     StartNewBlock();
@@ -2346,9 +2390,7 @@ private:
   void CalculateFlags_MUL(IR::OpSize SrcSize, Ref Res, Ref High);
   void CalculateFlags_UMUL(Ref High);
   void CalculateFlags_Logical(IR::OpSize SrcSize, Ref Res);
-  void CalculateFlags_ShiftLeft(IR::OpSize SrcSize, Ref Res, Ref Src1, Ref Src2);
   void CalculateFlags_ShiftLeftImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
-  void CalculateFlags_ShiftRight(IR::OpSize SrcSize, Ref Res, Ref Src1, Ref Src2);
   void CalculateFlags_ShiftRightImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
   void CalculateFlags_ShiftRightDoubleImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
   void CalculateFlags_ShiftRightImmediateCommon(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
@@ -2365,7 +2407,7 @@ private:
     LOGMAN_THROW_A_FMT(MMXState == MMXState_X87, "Expected state to be x87");
     _StackForceSlow();
     SetX87Top(Constant(0)); // top reset to zero
-    _StoreContext(OpSize::i8Bit, GPRClass, Constant(0xFFFFUL), offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+    _StoreContextGPR(OpSize::i8Bit, Constant(0xFFFFUL), offsetof(FEXCore::Core::CPUState, AbridgedFTW));
     MMXState = MMXState_MMX;
   }
 
@@ -2402,44 +2444,62 @@ private:
   IROp_IRHeader* CurrentHeader {};
 
   [[nodiscard]]
-  bool IsTSOEnabled(FEXCore::IR::RegisterClassType Class) const {
+  bool IsTSOEnabled(RegClass Class) const {
     if (ForceTSO == ForceTSOMode::ForceEnabled) {
       return true;
     } else if (ForceTSO == ForceTSOMode::ForceDisabled) {
       return false;
-    } else if (Class == FPRClass) {
+    } else if (Class == RegClass::FPR) {
       return CTX->IsVectorAtomicTSOEnabled();
     } else {
       return CTX->IsAtomicTSOEnabled();
     }
   }
 
-  Ref _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref Addr, Ref Value, IR::OpSize Align = IR::OpSize::i8Bit) {
+  Ref _StoreMemAutoTSO(RegClass Class, OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
     if (IsTSOEnabled(Class)) {
-      return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
+      return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
     } else {
-      return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
+      return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
     }
   }
-
-  Ref _LoadMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref ssa0, IR::OpSize Align = IR::OpSize::i8Bit) {
-    if (IsTSOEnabled(Class)) {
-      return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
-    } else {
-      return _LoadMem(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
-    }
+  Ref _StoreMemGPRAutoTSO(OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::GPR, Size, Addr, Value, Align);
+  }
+  Ref _StoreMemFPRAutoTSO(OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::FPR, Size, Addr, Value, Align);
   }
 
-  Ref _LoadMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
-    A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != GPRClass, Size);
+  Ref _LoadMemAutoTSO(RegClass Class, OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    if (IsTSOEnabled(Class)) {
+      return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+    } else {
+      return _LoadMem(Class, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+    }
+  }
+  Ref _LoadMemGPRAutoTSO(OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::GPR, Size, ssa0, Align);
+  }
+  Ref _LoadMemFPRAutoTSO(OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::FPR, Size, ssa0, Align);
+  }
+
+  Ref _LoadMemAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const auto B = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != RegClass::GPR, Size);
 
     if (AtomicTSO) {
-      return _LoadMemTSO(Class, Size, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _LoadMemTSO(Class, Size, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
     } else {
-      return _LoadMem(Class, Size, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _LoadMem(Class, Size, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
     }
   }
+  Ref _LoadMemGPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::GPR, Size, A, Align);
+  }
+  Ref _LoadMemFPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::FPR, Size, A, Align);
+  }
 
   AddressMode SelectPairAddressMode(AddressMode A, IR::OpSize Size) {
     LOGMAN_THROW_A_FMT(Size != IR::OpSize::iUnsized, "Invalid size!");
@@ -2457,56 +2517,72 @@ private:
   }
 
 
-  RefPair LoadMemPair(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref Base, unsigned Offset) {
+  RefPair LoadMemPair(RegClass Class, OpSize Size, Ref Base, uint32_t Offset) {
     RefPair Values = AllocatePair(Class, Size);
     _LoadMemPair(Class, Size, Base, Offset, Values.Low, Values.High);
     return Values;
   }
+  RefPair LoadMemPairFPR(OpSize Size, Ref Base, uint32_t Offset) {
+    return LoadMemPair(RegClass::FPR, Size, Base, Offset);
+  }
 
-  RefPair _LoadMemPairAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+  RefPair _LoadMemPairAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
 
     // Use ldp if possible, otherwise fallback on two loads.
-    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit & Size <= OpSize::i128Bit) {
-      A = SelectPairAddressMode(A, Size);
-      return LoadMemPair(Class, Size, A.Base, A.Offset);
-    } else {
-      AddressMode HighA = A;
-      HighA.Offset += 16;
-
-      return {
-        .Low = _LoadMemAutoTSO(Class, Size, A, Align),
-        .High = _LoadMemAutoTSO(Class, Size, HighA, Align),
-      };
+    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit && Size <= OpSize::i128Bit) {
+      const auto B = SelectPairAddressMode(A, Size);
+      return LoadMemPair(Class, Size, B.Base, B.Offset);
     }
+
+    AddressMode HighA = A;
+    HighA.Offset += 16;
+
+    return {
+      .Low = _LoadMemAutoTSO(Class, Size, A, Align),
+      .High = _LoadMemAutoTSO(Class, Size, HighA, Align),
+    };
+  }
+  RefPair _LoadMemPairFPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemPairAutoTSO(RegClass::FPR, Size, A, Align);
   }
 
-  Ref _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, Ref Value, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
-    A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != GPRClass, Size);
+  Ref _StoreMemAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const auto B = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != RegClass::GPR, Size);
 
     if (AtomicTSO) {
-      return _StoreMemTSO(Class, Size, Value, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _StoreMemTSO(Class, Size, Value, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
     } else {
-      return _StoreMem(Class, Size, Value, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _StoreMem(Class, Size, Value, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
     }
   }
+  Ref _StoreMemGPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::GPR, Size, A, Value, Align);
+  }
+  Ref _StoreMemFPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::FPR, Size, A, Value, Align);
+  }
 
-  void _StoreMemPairAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, Ref Value1, Ref Value2,
-                            IR::OpSize Align = IR::OpSize::i8Bit) {
+  void _StoreMemPairAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, Ref Value1, Ref Value2, OpSize Align = OpSize::i8Bit) {
     const auto SizeInt = IR::OpSizeToSize(Size);
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
 
     // Use stp if possible, otherwise fallback on two stores.
-    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit & Size <= OpSize::i128Bit) {
-      A = SelectPairAddressMode(A, Size);
-      _StoreMemPair(Class, Size, Value1, Value2, A.Base, A.Offset);
+    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit && Size <= OpSize::i128Bit) {
+      const auto B = SelectPairAddressMode(A, Size);
+      _StoreMemPair(Class, Size, Value1, Value2, B.Base, B.Offset);
     } else {
-      _StoreMemAutoTSO(Class, Size, A, Value1, OpSize::i8Bit);
-      A.Offset += SizeInt;
-      _StoreMemAutoTSO(Class, Size, A, Value2, OpSize::i8Bit);
+      auto B = A;
+
+      _StoreMemAutoTSO(Class, Size, B, Value1, OpSize::i8Bit);
+      B.Offset += SizeInt;
+      _StoreMemAutoTSO(Class, Size, B, Value2, OpSize::i8Bit);
     }
   }
+  void _StoreMemPairFPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value1, Ref Value2, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemPairAutoTSO(RegClass::FPR, Size, A, Value1, Value2, Align);
+  }
 
   Ref Pop(IR::OpSize Size, Ref SP_RMW) {
     Ref Value = _AllocateGPR(false);

FEXCore/Source/Interface/Core/OpcodeDispatcher/AVX_128.cpp

@@ -35,20 +35,16 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_LoadSource_WithOpSize(
   } else {
     LOGMAN_THROW_A_FMT(IsOperandMem(Operand, true), "only memory sources");
 
-    AddressMode A = DecodeAddress(Op, Operand, AccessType, true /* IsLoad */);
-
-    AddressMode HighA = A;
-    HighA.Offset += 16;
-
     if (Operand.IsSIB()) {
       const bool IsVSIB = (Op->Flags & X86Tables::DecodeFlags::FLAG_VSIB_BYTE) != 0;
       LOGMAN_THROW_A_FMT(!IsVSIB, "VSIB uses LoadVSIB instead");
     }
 
+    const AddressMode A = DecodeAddress(Op, Operand, AccessType, true /* IsLoad */);
     if (NeedsHigh) {
-      return _LoadMemPairAutoTSO(FPRClass, OpSize::i128Bit, A, OpSize::i8Bit);
+      return _LoadMemPairFPRAutoTSO(OpSize::i128Bit, A, OpSize::i8Bit);
     } else {
-      return {.Low = _LoadMemAutoTSO(FPRClass, OpSize::i128Bit, A, OpSize::i8Bit)};
+      return {.Low = _LoadMemFPRAutoTSO(OpSize::i128Bit, A, OpSize::i8Bit)};
     }
   }
 }
@@ -95,9 +91,9 @@ void OpDispatchBuilder::AVX128_StoreResult_WithOpSize(FEXCore::X86Tables::Decode
     AddressMode A = DecodeAddress(Op, Operand, AccessType, false /* IsLoad */);
 
     if (Src.High) {
-      _StoreMemPairAutoTSO(FPRClass, OpSize::i128Bit, A, Src.Low, Src.High, OpSize::i8Bit);
+      _StoreMemPairFPRAutoTSO(OpSize::i128Bit, A, Src.Low, Src.High, OpSize::i8Bit);
     } else {
-      _StoreMemAutoTSO(FPRClass, OpSize::i128Bit, A, Src.Low, OpSize::i8Bit);
+      _StoreMemFPRAutoTSO(OpSize::i128Bit, A, Src.Low, OpSize::i8Bit);
     }
   }
 }
@@ -151,13 +147,13 @@ void OpDispatchBuilder::AVX128_VMOVScalarImpl(OpcodeArgs, IR::OpSize ElementSize
     AVX128_StoreResult_WithOpSize(Op, Op->Dest, RefPair {.Low = Result, .High = High});
   } else if (Op->Dest.IsGPR()) {
     // VMOVSS/SD xmm1, mem32/mem64
-    Ref Src = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], ElementSize, Op->Flags);
+    Ref Src = LoadSourceFPR_WithOpSize(Op, Op->Src[1], ElementSize, Op->Flags);
     auto High = LoadZeroVector(OpSize::i128Bit);
     AVX128_StoreResult_WithOpSize(Op, Op->Dest, RefPair {.Low = Src, .High = High});
   } else {
     // VMOVSS/SD mem32/mem64, xmm1
     auto Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src.Low, ElementSize, OpSize::iInvalid);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src.Low, ElementSize);
   }
 }
 
@@ -351,7 +347,7 @@ void OpDispatchBuilder::AVX128_MOVVectorNT(OpcodeArgs) {
   if (Op->Dest.IsGPR()) {
     ///< MOVNTDQA load non-temporal comes from SSE4.1 and is extended by AVX/AVX2.
     RefPair Src {};
-    Ref SrcAddr = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, {.LoadData = false});
+    Ref SrcAddr = LoadSourceGPR(Op, Op->Src[0], Op->Flags, {.LoadData = false});
     Src.Low = _VLoadNonTemporal(OpSize::i128Bit, SrcAddr, 0);
 
     if (Is128Bit) {
@@ -362,7 +358,7 @@ void OpDispatchBuilder::AVX128_MOVVectorNT(OpcodeArgs) {
     AVX128_StoreResult_WithOpSize(Op, Op->Dest, Src);
   } else {
     auto Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, !Is128Bit, MemoryAccessType::STREAM);
-    Ref Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
+    Ref Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});
 
     if (Is128Bit) {
       // Single store non-temporal for 128-bit operations.
@@ -379,7 +375,7 @@ void OpDispatchBuilder::AVX128_MOVQ(OpcodeArgs) {
   if (Op->Src[0].IsGPR()) {
     Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
   } else {
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
   }
 
   // This instruction is a bit special that if the destination is a register then it'll ZEXT the 64bit source to 256bit
@@ -390,7 +386,7 @@ void OpDispatchBuilder::AVX128_MOVQ(OpcodeArgs) {
     Src.High = ZeroVector;
     AVX128_StoreResult_WithOpSize(Op, Op->Dest, Src);
   } else {
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src.Low, OpSize::i64Bit, OpSize::i64Bit);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src.Low, OpSize::i64Bit, OpSize::i64Bit);
   }
 }
 
@@ -399,7 +395,7 @@ void OpDispatchBuilder::AVX128_VMOVLP(OpcodeArgs) {
 
   if (!Op->Dest.IsGPR()) {
     ///< VMOVLPS/PD mem64, xmm1
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src1.Low, OpSize::i64Bit, OpSize::i64Bit);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src1.Low, OpSize::i64Bit, OpSize::i64Bit);
   } else if (!Op->Src[1].IsGPR()) {
     ///< VMOVLPS/PD xmm1, xmm2, mem64
     // Bits[63:0] come from Src2[63:0]
@@ -463,7 +459,7 @@ void OpDispatchBuilder::AVX128_VMOVDDUP(OpcodeArgs) {
     // 128-bit operation only loads 8-bytes.
     // 256-bit operation loads a full 32-bytes.
     if (Is128Bit) {
-      Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
+      Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
     } else {
       Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, true);
     }
@@ -558,18 +554,18 @@ void OpDispatchBuilder::AVX128_InsertCVTGPR_To_FPR(OpcodeArgs, IR::OpSize DstEle
 
   if (Op->Src[1].IsGPR()) {
     // If the source is a GPR then convert directly from the GPR.
-    auto Src2 = LoadSource_WithOpSize(GPRClass, Op, Op->Src[1], GetGPROpSize(), Op->Flags);
+    auto Src2 = LoadSourceGPR_WithOpSize(Op, Op->Src[1], GetGPROpSize(), Op->Flags);
     Result.Low = _VSToFGPRInsert(OpSize::i128Bit, DstElementSize, SrcSize, Src1.Low, Src2, false);
   } else if (SrcSize != DstElementSize) {
     // If the source is from memory but the Source size and destination size aren't the same,
     // then it is more optimal to load in to a GPR and convert between GPR->FPR.
     // ARM GPR->FPR conversion supports different size source and destinations while FPR->FPR doesn't.
-    auto Src2 = LoadSource(GPRClass, Op, Op->Src[1], Op->Flags);
+    auto Src2 = LoadSourceGPR(Op, Op->Src[1], Op->Flags);
     Result.Low = _VSToFGPRInsert(DstSize, DstElementSize, SrcSize, Src1.Low, Src2, false);
   } else {
     // In the case of cvtsi2s{s,d} where the source and destination are the same size,
     // then it is more optimal to load in to the FPR register directly and convert there.
-    auto Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+    auto Src2 = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
     // Always signed
     Result.Low = _VSToFVectorInsert(DstSize, DstElementSize, DstElementSize, Src1.Low, Src2, false, false);
   }
@@ -589,11 +585,11 @@ void OpDispatchBuilder::AVX128_CVTFPR_To_GPR(OpcodeArgs, IR::OpSize SrcElementSi
   if (Op->Src[0].IsGPR()) {
     Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
   } else {
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcElementSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcElementSize, Op->Flags);
   }
 
   Ref Result = CVTFPR_To_GPRImpl(Op, Src.Low, SrcElementSize, HostRoundingMode);
-  StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultGPR(Op, Result);
 }
 
 void OpDispatchBuilder::AVX128_VANDN(OpcodeArgs) {
@@ -636,7 +632,7 @@ void OpDispatchBuilder::AVX128_UCOMISx(OpcodeArgs, IR::OpSize ElementSize) {
   if (Op->Src[0].IsGPR()) {
     Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
   } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcSize, Op->Flags);
   }
 
   Comiss(ElementSize, Src1.Low, Src2.Low);
@@ -653,7 +649,7 @@ void OpDispatchBuilder::AVX128_VectorScalarInsertALU(OpcodeArgs, FEXCore::IR::IR
   if (Op->Src[1].IsGPR()) {
     Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
   } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
   }
 
   // If OpSize == ElementSize then it only does the lower scalar op
@@ -690,7 +686,7 @@ void OpDispatchBuilder::AVX128_InsertScalarFCMP(OpcodeArgs, IR::OpSize ElementSi
   if (Op->Src[1].IsGPR()) {
     Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
   } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
   }
 
   const uint8_t CompType = Op->Src[2].Literal();
@@ -708,12 +704,12 @@ void OpDispatchBuilder::AVX128_MOVBetweenGPR_FPR(OpcodeArgs) {
     RefPair Result {};
     if (Op->Src[0].IsGPR()) {
       // Loading from GPR and moving to Vector.
-      Ref Src = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], GetGPROpSize(), Op->Flags);
+      Ref Src = LoadSourceFPR_WithOpSize(Op, Op->Src[0], GetGPROpSize(), Op->Flags);
       // zext to 128bit
       Result.Low = _VCastFromGPR(OpSize::i128Bit, OpSizeFromSrc(Op), Src);
     } else {
       // Loading from Memory as a scalar. Zero extend
-      Result.Low = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Result.Low = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     }
 
     Result.High = LoadZeroVector(OpSize::i128Bit);
@@ -726,11 +722,11 @@ void OpDispatchBuilder::AVX128_MOVBetweenGPR_FPR(OpcodeArgs) {
       auto ElementSize = OpSizeFromDst(Op);
       // Extract element from GPR. Zero extending in the process.
       Src.Low = _VExtractToGPR(OpSizeFromSrc(Op), ElementSize, Src.Low, 0);
-      StoreResult(GPRClass, Op, Op->Dest, Src.Low, OpSize::iInvalid);
+      StoreResultGPR(Op, Op->Dest, Src.Low);
     } else {
       // Storing first element to memory.
-      Ref Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
-      _StoreMem(FPRClass, OpSizeFromDst(Op), Dest, Src.Low, OpSize::i8Bit);
+      Ref Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});
+      _StoreMemFPR(OpSizeFromDst(Op), Dest, Src.Low, OpSize::i8Bit);
     }
   }
 }
@@ -758,7 +754,7 @@ void OpDispatchBuilder::AVX128_PExtr(OpcodeArgs, IR::OpSize ElementSize) {
     const auto GPRSize = GetGPROpSize();
     // Extract already zero extends the result.
     Ref Result = _VExtractToGPR(OpSize::i128Bit, OverridenElementSize, Src.Low, Index);
-    StoreResult_WithOpSize(GPRClass, Op, Op->Dest, Result, GPRSize, OpSize::iInvalid);
+    StoreResultGPR_WithOpSize(Op, Op->Dest, Result, GPRSize);
     return;
   }
 
@@ -779,7 +775,7 @@ void OpDispatchBuilder::AVX128_ExtendVectorElements(OpcodeArgs, IR::OpSize Eleme
       const auto SrcSize = OpSizeFromSrc(Op);
       const auto LoadSize = Is256Bit ? IR::SizeToOpSize(IR::OpSizeToSize(SrcSize) * 2) : SrcSize;
 
-      return LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags);
+      return LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags);
     }
   };
 
@@ -868,7 +864,7 @@ void OpDispatchBuilder::AVX128_MOVMSK(OpcodeArgs, IR::OpSize ElementSize) {
     auto GPRHigh = Mask8Byte(Src.High);
     GPR = _Orlshl(OpSize::i64Bit, GPRLow, GPRHigh, 2);
   }
-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, GPR, GetGPROpSize(), OpSize::iInvalid);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, GPR, GetGPROpSize());
 }
 
 void OpDispatchBuilder::AVX128_MOVMSKB(OpcodeArgs) {
@@ -897,7 +893,7 @@ void OpDispatchBuilder::AVX128_MOVMSKB(OpcodeArgs) {
     Result = _Orlshl(OpSize::i64Bit, Result, ResultHigh, 16);
   }
 
-  StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultGPR(Op, Result);
 }
 
 void OpDispatchBuilder::AVX128_PINSRImpl(OpcodeArgs, IR::OpSize ElementSize, const X86Tables::DecodedOperand& Src1Op,
@@ -910,7 +906,7 @@ void OpDispatchBuilder::AVX128_PINSRImpl(OpcodeArgs, IR::OpSize ElementSize, con
 
   if (Src2Op.IsGPR()) {
     // If the source is a GPR then convert directly from the GPR.
-    auto Src2 = LoadSource_WithOpSize(GPRClass, Op, Src2Op, GetGPROpSize(), Op->Flags);
+    auto Src2 = LoadSourceGPR_WithOpSize(Op, Src2Op, GetGPROpSize(), Op->Flags);
     Result.Low = _VInsGPR(OpSize::i128Bit, ElementSize, Index, Src1.Low, Src2);
   } else {
     // If loading from memory then we only load the element size
@@ -1047,7 +1043,7 @@ void OpDispatchBuilder::AVX128_InsertScalar_CVT_Float_To_Float(OpcodeArgs, IR::O
   // Then zero extends the top 128-bit.
   const auto SrcSize = Op->Src[1].IsGPR() ? OpSize::i128Bit : SrcElementSize;
   auto Src1 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
-  Ref Src2 = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags, {.AllowUpperGarbage = true});
+  Ref Src2 = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags, {.AllowUpperGarbage = true});
 
   Ref Result = _VFToFScalarInsert(OpSize::i128Bit, DstElementSize, SrcElementSize, Src1.Low, Src2, false);
   AVX128_StoreResult_WithOpSize(Op, Op->Dest, AVX128_Zext(Result));
@@ -1076,7 +1072,7 @@ void OpDispatchBuilder::AVX128_Vector_CVT_Float_To_Float(OpcodeArgs, IR::OpSize
   } else {
     // Handle 64-bit memory source.
     // In the case of cvtps2pd xmm, m64.
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags);
   }
 
   RefPair Result {};
@@ -1154,7 +1150,7 @@ void OpDispatchBuilder::AVX128_Vector_CVT_Int_To_Float(OpcodeArgs, IR::OpSize Sr
       // unnecessarily zero extend the vector. Otherwise, if
       // memory, then we want to load the element size exactly.
       const auto LoadSize = IR::SizeToOpSize(8 * (IR::OpSizeToSize(Size) / 16));
-      return RefPair {.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags)};
+      return RefPair {.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags)};
     } else {
       return AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, !Is128Bit);
     }
@@ -1304,7 +1300,7 @@ void OpDispatchBuilder::AVX128_InsertScalarRound(OpcodeArgs, IR::OpSize ElementS
   if (Op->Src[1].IsGPR()) {
     Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
   } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
   }
 
   // If OpSize == ElementSize then it only does the lower scalar op
@@ -1573,20 +1569,20 @@ void OpDispatchBuilder::AVX128_VMASKMOVImpl(OpcodeArgs, IR::OpSize ElementSize,
     auto Address = MakeAddress(Op->Dest);
 
     auto Data = AVX128_LoadSource_WithOpSize(Op, DataOp, Op->Flags, !Is128Bit);
-    _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Data.Low, Address, Invalid(), MEM_OFFSET_SXTX, 1);
+    _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Data.Low, Address, Invalid(), MemOffsetType::SXTX, 1);
     if (!Is128Bit) {
-      _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Data.High, Address, _InlineConstant(16), MEM_OFFSET_SXTX, 1);
+      _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Data.High, Address, _InlineConstant(16), MemOffsetType::SXTX, 1);
     }
   } else {
     auto Address = MakeAddress(DataOp);
 
     RefPair Result {};
-    Result.Low = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Address, Invalid(), MEM_OFFSET_SXTX, 1);
+    Result.Low = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Address, Invalid(), MemOffsetType::SXTX, 1);
 
     if (Is128Bit) {
       Result.High = LoadZeroVector(OpSize::i128Bit);
     } else {
-      Result.High = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Address, _InlineConstant(16), MEM_OFFSET_SXTX, 1);
+      Result.High = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Address, _InlineConstant(16), MemOffsetType::SXTX, 1);
     }
     AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
   }
@@ -1616,11 +1612,11 @@ void OpDispatchBuilder::AVX128_MASKMOV(OpcodeArgs) {
   // RDI source (DS prefix by default)
   auto MemDest = MakeSegmentAddress(X86State::REG_RDI, Op->Flags, X86Tables::DecodeFlags::FLAG_DS_PREFIX);
 
-  Ref XMMReg = _LoadMem(FPRClass, Size, MemDest, OpSize::i8Bit);
+  Ref XMMReg = _LoadMemFPR(Size, MemDest, OpSize::i8Bit);
 
   // If the Mask element high bit is set then overwrite the element with the source, else keep the memory variant
   XMMReg = _VBSL(Size, MaskSrc.Low, VectorSrc.Low, XMMReg);
-  _StoreMem(FPRClass, Size, MemDest, XMMReg, OpSize::i8Bit);
+  _StoreMemFPR(Size, MemDest, XMMReg, OpSize::i8Bit);
 }
 
 void OpDispatchBuilder::AVX128_VectorVariableBlend(OpcodeArgs, IR::OpSize ElementSize) {
@@ -1660,7 +1656,7 @@ void OpDispatchBuilder::AVX128_SaveAVXState(Ref MemBase) {
 
   for (uint32_t i = 0; i < NumRegs; i += 2) {
     RefPair Pair = LoadContextPair(OpSize::i128Bit, AVXHigh0Index + i);
-    _StoreMemPair(FPRClass, OpSize::i128Bit, Pair.Low, Pair.High, MemBase, i * 16 + 576);
+    _StoreMemPairFPR(OpSize::i128Bit, Pair.Low, Pair.High, MemBase, i * 16 + 576);
   }
 }
 
@@ -1668,7 +1664,7 @@ void OpDispatchBuilder::AVX128_RestoreAVXState(Ref MemBase) {
   const auto NumRegs = Is64BitMode ? 16U : 8U;
 
   for (uint32_t i = 0; i < NumRegs; i += 2) {
-    auto YMMHRegs = LoadMemPair(FPRClass, OpSize::i128Bit, MemBase, i * 16 + 576);
+    auto YMMHRegs = LoadMemPairFPR(OpSize::i128Bit, MemBase, i * 16 + 576);
 
     AVX128_StoreXMMRegister(i, YMMHRegs.Low, true);
     AVX128_StoreXMMRegister(i + 1, YMMHRegs.High, true);
@@ -1960,7 +1956,7 @@ void OpDispatchBuilder::AVX128_VFMAImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx,
 }
 
 void OpDispatchBuilder::AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx) {
-  const auto SrcSize = OpSizeFromSrc(Op);
+  const OpSize ElementSize = Op->Flags & X86Tables::DecodeFlags::FLAG_OPTION_AVX_W ? OpSize::i64Bit : OpSize::i32Bit;
 
   auto Dest = AVX128_LoadSource_WithOpSize(Op, Op->Dest, Op->Flags, false).Low;
   auto Src1 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false).Low;
@@ -1968,13 +1964,13 @@ void OpDispatchBuilder::AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Sr
   if (Op->Src[1].IsGPR()) {
     Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false).Low;
   } else {
-    Src2 = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2 = LoadSourceFPR_WithOpSize(Op, Op->Src[1], ElementSize, Op->Flags);
   }
 
   Ref Sources[3] = {Dest, Src1, Src2};
 
   DeriveOp(Result_Low, IROp,
-           _VFMLAScalarInsert(OpSize::i128Bit, SrcSize, Dest, Sources[Src1Idx - 1], Sources[Src2Idx - 1], Sources[AddendIdx - 1]));
+           _VFMLAScalarInsert(OpSize::i128Bit, ElementSize, Dest, Sources[Src1Idx - 1], Sources[Src2Idx - 1], Sources[AddendIdx - 1]));
   AVX128_StoreResult_WithOpSize(Op, Op->Dest, AVX128_Zext(Result_Low));
 }
 
@@ -2016,8 +2012,8 @@ void OpDispatchBuilder::AVX128_VFMAddSubImpl(OpcodeArgs, bool AddSub, uint8_t Sr
   AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
 }
 
-OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest,
-                                                                  RefPair Mask, RefVSIB VSIB) {
+OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpcodeArgs, OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize,
+                                                                  RefPair Dest, RefPair Mask, RefVSIB VSIB) {
   LOGMAN_THROW_A_FMT(AddrElementSize == OpSize::i32Bit || AddrElementSize == OpSize::i64Bit, "Unknown address element size");
   const auto Is128Bit = Size == OpSize::i128Bit;
 
@@ -2061,10 +2057,13 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O
     }
   }
 
+  const auto GPRSize = GetGPROpSize();
+  auto AddrSize = (Op->Flags & X86Tables::DecodeFlags::FLAG_ADDRESS_SIZE) != 0 ? (GPRSize >> 1) : GPRSize;
+
   RefPair Result {};
   ///< Calculate the low-half.
   Result.Low = _VLoadVectorGatherMasked(OpSize::i128Bit, ElementLoadSize, Dest.Low, Mask.Low, BaseAddr, VSIB.Low, VSIB.High,
-                                        AddrElementSize, VSIB.Scale, 0, 0);
+                                        AddrElementSize, VSIB.Scale, 0, 0, AddrSize);
 
   if (Is128Bit) {
     Result.High = LoadZeroVector(OpSize::i128Bit);
@@ -2101,7 +2100,7 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O
 
     ///< Calculate the high-half.
     auto ResultHigh = _VLoadVectorGatherMasked(OpSize::i128Bit, ElementLoadSize, DestReg, MaskReg, BaseAddr, AddrAddressing.Low,
-                                               AddrAddressing.High, AddrElementSize, VSIB.Scale, DataElementOffset, IndexElementOffset);
+                                               AddrAddressing.High, AddrElementSize, VSIB.Scale, DataElementOffset, IndexElementOffset, AddrSize);
 
     if (AddrElementSize == OpSize::i64Bit && ElementLoadSize == OpSize::i32Bit) {
       // If we only fetched 128-bits worth of data then the upper-result is all zero.
@@ -2114,7 +2113,7 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O
   return Result;
 }
 
-OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(Ref Dest, Ref Mask, RefVSIB VSIB) {
+OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(OpcodeArgs, Ref Dest, Ref Mask, RefVSIB VSIB) {
 
   ///< BaseAddr doesn't need to exist, calculate that here.
   Ref BaseAddr = VSIB.BaseAddr;
@@ -2142,8 +2141,11 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(Ref Dest, R
 
   RefPair Result {};
 
+  const auto GPRSize = GetGPROpSize();
+  auto AddrSize = (Op->Flags & X86Tables::DecodeFlags::FLAG_ADDRESS_SIZE) != 0 ? (GPRSize >> 1) : GPRSize;
+
   ///< Calculate the low-half.
-  Result.Low = _VLoadVectorGatherMaskedQPS(OpSize::i128Bit, OpSize::i32Bit, Dest, Mask, BaseAddr, VSIB.Low, VSIB.High, VSIB.Scale);
+  Result.Low = _VLoadVectorGatherMaskedQPS(OpSize::i128Bit, OpSize::i32Bit, Dest, Mask, BaseAddr, VSIB.Low, VSIB.High, VSIB.Scale, AddrSize);
   Result.High = LoadZeroVector(OpSize::i128Bit);
   if (VSIB.High == Invalid()) {
     // Special case for only loading two floats.
@@ -2202,15 +2204,15 @@ void OpDispatchBuilder::AVX128_VPGATHER(OpcodeArgs, OpSize AddrElementSize) {
     }
 
     ///< AddressElementSize is now OpSize::i64Bit
-    Result = AVX128_VPGatherQPSImpl(Dest.Low, Mask.Low, VSIBLow);
+    Result = AVX128_VPGatherQPSImpl(Op, Dest.Low, Mask.Low, VSIBLow);
     if (NeedsHighAddrBytes) {
-      auto Res = AVX128_VPGatherQPSImpl(Dest.High, Mask.High, VSIBHigh);
+      auto Res = AVX128_VPGatherQPSImpl(Op, Dest.High, Mask.High, VSIBHigh);
       Result.High = Res.Low;
     }
   } else if (AddrElementSize == OpSize::i64Bit && ElementLoadSize == OpSize::i32Bit) {
-    Result = AVX128_VPGatherQPSImpl(Dest.Low, Mask.Low, VSIB);
+    Result = AVX128_VPGatherQPSImpl(Op, Dest.Low, Mask.Low, VSIB);
   } else {
-    Result = AVX128_VPGatherImpl(Size, ElementLoadSize, AddrElementSize, Dest, Mask, VSIB);
+    Result = AVX128_VPGatherImpl(Op, Size, ElementLoadSize, AddrElementSize, Dest, Mask, VSIB);
   }
   AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
 
@@ -2234,7 +2236,7 @@ void OpDispatchBuilder::AVX128_VCVTPH2PS(OpcodeArgs) {
     // In the event that a memory operand is used as the source operand,
     // the access width will always be half the size of the destination vector width
     // (i.e. 128-bit vector -> 64-bit mem, 256-bit vector -> 128-bit mem)
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcSize, Op->Flags);
   }
 
   RefPair Result {};
@@ -2289,7 +2291,7 @@ void OpDispatchBuilder::AVX128_VCVTPS2PH(OpcodeArgs) {
   }
 
   if (!Op->Dest.IsGPR()) {
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Result.Low, StoreSize, OpSize::iInvalid);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Result.Low, StoreSize);
   } else {
     AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
   }

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

@@ -23,8 +23,8 @@ void OpDispatchBuilder::SHA1NEXTEOp(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
 
   // ARMv8 SHA1 extension provides a `SHA1H` instruction which does a fixed rotate by 30.
   // This only operates on element 0 rather than element 3. We don't have the luxury of rewriting the x86 SHA algorithm to take advantage of this.
@@ -36,7 +36,7 @@ void OpDispatchBuilder::SHA1NEXTEOp(OpcodeArgs) {
   auto Tmp = _VAdd(OpSize::i128Bit, OpSize::i32Bit, Src, RotatedNode);
   auto Result = _VInsElement(OpSize::i128Bit, OpSize::i32Bit, 3, 3, Src, Tmp);
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::SHA1MSG1Op(OpcodeArgs) {
@@ -44,15 +44,15 @@ void OpDispatchBuilder::SHA1MSG1Op(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
 
   Ref NewVec = _VExtr(OpSize::i128Bit, OpSize::i64Bit, Dest, Src, 1);
 
   // [W0, W1, W2, W3] ^ [W2, W3, W4, W5]
   Ref Result = _VXor(OpSize::i128Bit, OpSize::i8Bit, Dest, NewVec);
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
@@ -60,8 +60,8 @@ void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
 
   // ARM SHA1 mostly matches x86 semantics, except the input and outputs are both flipped from elements 0,1,2,3 to 3,2,1,0.
   auto Src1 = SHADataShuffle(Dest);
@@ -70,7 +70,7 @@ void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
   // The result is swizzled differently than expected
   auto Result = SHADataShuffle(_VSha1SU1(Src1, Src2));
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
@@ -79,8 +79,8 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
     return;
   }
   const uint64_t Imm8 = Op->Src[1].Literal() & 0b11;
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
 
   Ref Result {};
   Ref ConstantVector {};
@@ -112,7 +112,7 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
   case 3: Result = SHADataShuffle(_VSha1P(Src1, ZeroRegister, Src2)); break;
   }
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::SHA256MSG1Op(OpcodeArgs) {
@@ -120,12 +120,12 @@ void OpDispatchBuilder::SHA256MSG1Op(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
 
   auto Result = _VSha256U0(Dest, Src);
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {
@@ -133,8 +133,8 @@ void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
 
   auto Src1 = _VExtr(OpSize::i128Bit, OpSize::i32Bit, Dest, Dest, 3);
   auto DupDst = _VDupElement(OpSize::i128Bit, OpSize::i32Bit, Dest, 3);
@@ -142,7 +142,7 @@ void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {
 
   auto Result = _VSha256U1(Src1, Src2);
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
@@ -150,8 +150,8 @@ void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   // Hardcoded to XMM0
   auto XMM0 = LoadXMMRegister(0);
 
@@ -177,7 +177,7 @@ void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
   auto B = _VSha256H2(EFGH, ABCD, Key);
   auto Result = shuffle_abcd(A, B);
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
@@ -185,9 +185,9 @@ void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   Ref Result = _VAESImc(Src);
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::AESEncOp(OpcodeArgs) {
@@ -195,10 +195,10 @@ void OpDispatchBuilder::AESEncOp(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   Ref Result = _VAESEnc(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::VAESEncOp(OpcodeArgs) {
@@ -208,11 +208,11 @@ void OpDispatchBuilder::VAESEncOp(OpcodeArgs) {
   // TODO: Handle 256-bit VAESENC.
   LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESENC unimplemented");
 
-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
   Ref Result = _VAESEnc(DstSize, State, Key, LoadZeroVector(DstSize));
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::AESEncLastOp(OpcodeArgs) {
@@ -220,10 +220,10 @@ void OpDispatchBuilder::AESEncLastOp(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   Ref Result = _VAESEncLast(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::VAESEncLastOp(OpcodeArgs) {
@@ -233,11 +233,11 @@ void OpDispatchBuilder::VAESEncLastOp(OpcodeArgs) {
   // TODO: Handle 256-bit VAESENCLAST.
   LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESENCLAST unimplemented");
 
-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
   Ref Result = _VAESEncLast(DstSize, State, Key, LoadZeroVector(DstSize));
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::AESDecOp(OpcodeArgs) {
@@ -245,10 +245,10 @@ void OpDispatchBuilder::AESDecOp(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   Ref Result = _VAESDec(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::VAESDecOp(OpcodeArgs) {
@@ -258,11 +258,11 @@ void OpDispatchBuilder::VAESDecOp(OpcodeArgs) {
   // TODO: Handle 256-bit VAESDEC.
   LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESDEC unimplemented");
 
-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
   Ref Result = _VAESDec(DstSize, State, Key, LoadZeroVector(DstSize));
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::AESDecLastOp(OpcodeArgs) {
@@ -270,10 +270,10 @@ void OpDispatchBuilder::AESDecLastOp(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   Ref Result = _VAESDecLast(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::VAESDecLastOp(OpcodeArgs) {
@@ -283,15 +283,15 @@ void OpDispatchBuilder::VAESDecLastOp(OpcodeArgs) {
   // TODO: Handle 256-bit VAESDECLAST.
   LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESDECLAST unimplemented");
 
-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
   Ref Result = _VAESDecLast(DstSize, State, Key, LoadZeroVector(DstSize));
 
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 Ref OpDispatchBuilder::AESKeyGenAssistImpl(OpcodeArgs) {
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   const uint64_t RCON = Op->Src[1].Literal();
 
   auto KeyGenSwizzle = LoadAndCacheNamedVectorConstant(OpSize::i128Bit, NAMED_VECTOR_AESKEYGENASSIST_SWIZZLE);
@@ -305,7 +305,7 @@ void OpDispatchBuilder::AESKeyGenAssist(OpcodeArgs) {
   }
 
   Ref Result = AESKeyGenAssistImpl(Op);
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }
 
 void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
@@ -313,12 +313,12 @@ void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
     UnimplementedOp(Op);
     return;
   }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   const auto Selector = static_cast<uint8_t>(Op->Src[1].Literal());
 
   auto Res = _PCLMUL(OpSize::i128Bit, Dest, Src, Selector & 0b1'0001);
-  StoreResult(FPRClass, Op, Res, OpSize::iInvalid);
+  StoreResultFPR(Op, Res);
 }
 
 void OpDispatchBuilder::VPCLMULQDQOp(OpcodeArgs) {
@@ -328,12 +328,12 @@ void OpDispatchBuilder::VPCLMULQDQOp(OpcodeArgs) {
   }
   const auto DstSize = OpSizeFromDst(Op);
 
-  Ref Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref Src1 = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Src2 = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
   const auto Selector = static_cast<uint8_t>(Op->Src[2].Literal());
 
   Ref Res = _PCLMUL(DstSize, Src1, Src2, Selector & 0b1'0001);
-  StoreResult(FPRClass, Op, Res, OpSize::iInvalid);
+  StoreResultFPR(Op, Res);
 }
 
 } // namespace FEXCore::IR

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

@@ -263,7 +263,7 @@ void OpDispatchBuilder::CalculateDeferredFlags() {
 Ref OpDispatchBuilder::IncrementByCarry(OpSize OpSize, Ref Src) {
   // If CF not inverted, we use .cc since the increment happens when the
   // condition is false. If CF inverted, invert to use .cs. A bit mindbendy.
-  return _NZCVSelectIncrement(OpSize, {CFInverted ? COND_UGE : COND_ULT}, Src, Src);
+  return _NZCVSelectIncrement(OpSize, CFInverted ? CondClass::UGE : CondClass::ULT, Src, Src);
 }
 
 Ref OpDispatchBuilder::CalculateFlags_ADC(IR::OpSize SrcSize, Ref Src1, Ref Src2) {
@@ -290,7 +290,7 @@ Ref OpDispatchBuilder::CalculateFlags_ADC(IR::OpSize SrcSize, Ref Src1, Ref Src2
     Res = _Bfe(OpSize, IR::OpSizeAsBits(SrcSize), 0, Res);
 
     // TODO: We can fold that second Bfe in (cmp uxth).
-    auto SelectCFInv = Select01(OpSize, CondClassType {COND_UGE}, Res, Src2PlusCF);
+    auto SelectCFInv = Select01(OpSize, CondClass::UGE, Res, Src2PlusCF);
 
     SetNZ_ZeroCV(SrcSize, Res);
     SetCFInverted(SelectCFInv);
@@ -324,7 +324,7 @@ Ref OpDispatchBuilder::CalculateFlags_SBB(IR::OpSize SrcSize, Ref Src1, Ref Src2
     Res = Sub(OpSize, Src1, Src2PlusCF);
     Res = _Bfe(OpSize, IR::OpSizeAsBits(SrcSize), 0, Res);
 
-    auto SelectCFInv = Select01(OpSize, CondClassType {COND_UGE}, Src1, Src2PlusCF);
+    auto SelectCFInv = Select01(OpSize, CondClass::UGE, Src1, Src2PlusCF);
 
     SetNZ_ZeroCV(SrcSize, Res);
     SetCFInverted(SelectCFInv);
@@ -406,7 +406,7 @@ void OpDispatchBuilder::CalculateFlags_MUL(IR::OpSize SrcSize, Ref Res, Ref High
   // If High = SignBit, then sets to nZCv. Else sets to nzcV. Since SF/ZF
   // undefined, this does what we need after inverting carry.
   auto Zero = _InlineConstant(0);
-  _CondSubNZCV(OpSize::i64Bit, Zero, Zero, CondClassType {COND_EQ}, 0x1 /* nzcV */);
+  _CondSubNZCV(OpSize::i64Bit, Zero, Zero, CondClass::EQ, 0x1 /* nzcV */);
   CFInverted = true;
 }
 
@@ -423,7 +423,7 @@ void OpDispatchBuilder::CalculateFlags_UMUL(Ref High) {
 
   // If High = 0, then sets to nZCv. Else sets to nzcV. Since SF/ZF undefined,
   // this does what we need.
-  _CondSubNZCV(Size, Zero, Zero, CondClassType {COND_EQ}, 0x1 /* nzcV */);
+  _CondSubNZCV(Size, Zero, Zero, CondClass::EQ, 0x1 /* nzcV */);
   CFInverted = true;
 }
 

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

@@ -151,6 +151,9 @@ constexpr DispatchTableEntry OpDispatch_SecondaryGroupTables[] = {
   {OPD(FEXCore::X86Tables::TYPE_GROUP_16, PF_F2, 3), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, false, false, 3>},
   {OPD(FEXCore::X86Tables::TYPE_GROUP_16, PF_F2, 4), 4, &OpDispatchBuilder::NOPOp},
 
+  // GROUP 17
+  {OPD(FEXCore::X86Tables::TYPE_GROUP_17, PF_66, 0), 1, &OpDispatchBuilder::Extrq_imm},
+
   // GROUP P
   {OPD(FEXCore::X86Tables::TYPE_GROUP_P, PF_NONE, 0), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, false, false, 1>},
   {OPD(FEXCore::X86Tables::TYPE_GROUP_P, PF_NONE, 1), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, true, false, 1>},

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

@@ -145,7 +145,7 @@ constexpr DispatchTableEntry OpDispatch_TwoByteOpTable[] = {
 
 #ifndef _WIN32
   // FEX reserved instructions
-  {0x37, 1, &OpDispatchBuilder::CallbackReturnOp},
+  {0x3E, 1, &OpDispatchBuilder::CallbackReturnOp},
   {0x3F, 1, &OpDispatchBuilder::ThunkOp},
 #endif
 };
@@ -198,6 +198,8 @@ constexpr DispatchTableEntry OpDispatch_SecondaryRepNEModTables[] = {
   {0x5E, 1, &OpDispatchBuilder::VectorScalarInsertALUOp<IR::OP_VFDIVSCALARINSERT, OpSize::i64Bit>},
   {0x5F, 1, &OpDispatchBuilder::VectorScalarInsertALUOp<IR::OP_VFMAXSCALARINSERT, OpSize::i64Bit>},
   {0x70, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::PSHUFWOp, true>},
+  {0x78, 1, &OpDispatchBuilder::Insertq_imm},
+  {0x79, 1, &OpDispatchBuilder::Insertq},
   {0x7C, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VFADDP, OpSize::i32Bit>},
   {0x7D, 1, &OpDispatchBuilder::HSUBP<OpSize::i32Bit>},
   {0xD0, 1, &OpDispatchBuilder::ADDSUBPOp<OpSize::i32Bit>},
@@ -256,6 +258,7 @@ constexpr DispatchTableEntry OpDispatch_SecondaryOpSizeModTables[] = {
   {0x75, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VCMPEQ, OpSize::i16Bit>},
   {0x76, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VCMPEQ, OpSize::i32Bit>},
   {0x78, 1, nullptr}, // GROUP 17
+  {0x79, 1, &OpDispatchBuilder::Extrq},
   {0x7C, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VFADDP, OpSize::i64Bit>},
   {0x7D, 1, &OpDispatchBuilder::HSUBP<OpSize::i64Bit>},
   {0x7E, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::MOVBetweenGPR_FPR, OpDispatchBuilder::VectorOpType::SSE>},

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

@@ -17,7 +17,6 @@ $end_info$
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/FPState.h>
 
-#include <cmath>
 #include <stddef.h>
 #include <stdint.h>
 
@@ -28,7 +27,7 @@ class OrderedNode;
 Ref OpDispatchBuilder::GetX87Top() {
   // Yes, we are storing 3 bits in a single flag register.
   // Deal with it
-  return _LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
+  return _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
 }
 
 void OpDispatchBuilder::SetX87FTW(Ref FTW) {
@@ -52,24 +51,22 @@ void OpDispatchBuilder::SetX87FTW(Ref FTW) {
   FTW = _Orlshr(OpSize::i32Bit, FTW, FTW, 4);
 
   // ...and that's it. StoreContext implicitly does the final masking.
-  _StoreContext(OpSize::i8Bit, GPRClass, FTW, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+  _StoreContextGPR(OpSize::i8Bit, FTW, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
 }
 
 void OpDispatchBuilder::SetX87Top(Ref Value) {
-  _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
+  _StoreContextGPR(OpSize::i8Bit, Value, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
 }
 
 // Float LoaD operation with memory operand
 void OpDispatchBuilder::FLD(OpcodeArgs, IR::OpSize Width) {
-  const auto ReadWidth = (Width == OpSize::f80Bit) ? OpSize::i128Bit : Width;
-
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], Width, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], Width, Op->Flags);
   Ref ConvertedData = Data;
   // Convert to 80bit float
   if (Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
-    ConvertedData = _F80CVTTo(Data, ReadWidth);
+    ConvertedData = _F80CVTTo(Data, Width);
   }
-  _PushStack(ConvertedData, Data, ReadWidth, true);
+  _PushStack(ConvertedData, Data, Width);
 }
 
 // Float LoaD operation with memory operand
@@ -79,27 +76,27 @@ void OpDispatchBuilder::FLDFromStack(OpcodeArgs) {
 
 void OpDispatchBuilder::FBLD(OpcodeArgs) {
   // Read from memory
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
   Ref ConvertedData = _F80BCDLoad(Data);
-  _PushStack(ConvertedData, Data, OpSize::i128Bit, true);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }
 
 void OpDispatchBuilder::FBSTP(OpcodeArgs) {
   Ref converted = _F80BCDStore(_ReadStackValue(0));
-  StoreResult_WithOpSize(FPRClass, Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
+  StoreResultFPR_WithOpSize(Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
   _PopStackDestroy();
 }
 
 void OpDispatchBuilder::FLD_Const(OpcodeArgs, NamedVectorConstant K) {
   // Update TOP
   Ref Data = LoadAndCacheNamedVectorConstant(OpSize::i128Bit, K);
-  _PushStack(Data, Data, OpSize::i128Bit, true);
+  _PushStack(Data, Data, OpSize::f80Bit);
 }
 
 void OpDispatchBuilder::FILD(OpcodeArgs) {
   const auto ReadWidth = OpSizeFromSrc(Op);
   // Read from memory
-  Ref Data = LoadSource_WithOpSize(GPRClass, Op, Op->Src[0], ReadWidth, Op->Flags);
+  Ref Data = LoadSourceGPR_WithOpSize(Op, Op->Src[0], ReadWidth, Op->Flags);
 
   // Sign extend to 64bits
   if (ReadWidth != OpSize::i64Bit) {
@@ -112,27 +109,28 @@ void OpDispatchBuilder::FILD(OpcodeArgs) {
   // Extract sign and make integer absolute
   auto zero = Constant(0);
   _SubNZCV(OpSize::i64Bit, Data, zero);
-  auto sign = _NZCVSelect(OpSize::i64Bit, CondClassType {COND_SLT}, Constant(0x8000), zero);
-  auto absolute = _Neg(OpSize::i64Bit, Data, CondClassType {COND_MI});
+  auto sign = _NZCVSelect(OpSize::i64Bit, CondClass::SLT, Constant(0x8000), zero);
+  auto absolute = _Neg(OpSize::i64Bit, Data, CondClass::MI);
 
   // left justify the absolute integer
   auto shift = Sub(OpSize::i64Bit, Constant(63), _FindMSB(IR::OpSize::i64Bit, absolute));
   auto shifted = _Lshl(OpSize::i64Bit, absolute, shift);
 
   auto adjusted_exponent = Sub(OpSize::i64Bit, Constant(0x3fff + 63), shift);
-  auto zeroed_exponent = _Select(COND_EQ, absolute, zero, zero, adjusted_exponent);
+  auto zeroed_exponent = _Select(OpSize::i64Bit, OpSize::i64Bit, CondClass::EQ, absolute, zero, zero, adjusted_exponent);
   auto upper = _Or(OpSize::i64Bit, sign, zeroed_exponent);
 
   Ref ConvertedData = _VLoadTwoGPRs(shifted, upper);
-  _PushStack(ConvertedData, Data, ReadWidth, false);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }
 
 void OpDispatchBuilder::FST(OpcodeArgs, IR::OpSize Width) {
-  const auto SourceSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
+  LOGMAN_THROW_A_FMT(Width == OpSize::i32Bit || Width == OpSize::i64Bit || Width == OpSize::f80Bit, "Invalid store width for FST");
+  const auto SourceSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::f80Bit;
   AddressMode A = DecodeAddress(Op, Op->Dest, MemoryAccessType::DEFAULT, false);
 
   A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, false, false, Width);
-  _StoreStackMem(SourceSize, Width, A.Base, A.Index, OpSize::iInvalid, A.IndexType, A.IndexScale, /*Float=*/true);
+  _StoreStackMem(SourceSize, Width, A.Base, A.Index, OpSize::iInvalid, A.IndexType, A.IndexScale);
 
   if (Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) {
     _PopStackDestroy();
@@ -166,12 +164,12 @@ void OpDispatchBuilder::FIST(OpcodeArgs, bool Truncate) {
     // Check for NaN/Infinity: exponent = 0x7fff
     SaveNZCV();
     _TestNZ(OpSize::i64Bit, Exponent, Constant(0x7fff));
-    Ref IsSpecial = _NZCVSelect01({COND_EQ});
+    Ref IsSpecial = _NZCVSelect01(CondClass::EQ);
 
     // For overflow detection, check if exponent indicates a value >= 2^15
     // Biased exponent for 2^15 is 0x3fff + 15 = 0x400e
     SubWithFlags(OpSize::i64Bit, Exponent, 0x400e);
-    Ref IsOverflow = _NZCVSelect01({COND_UGE});
+    Ref IsOverflow = _NZCVSelect01(CondClass::UGE);
 
     // Set Invalid Operation flag if overflow or special value
     Ref InvalidFlag = _Or(OpSize::i64Bit, IsSpecial, IsOverflow);
@@ -180,7 +178,7 @@ void OpDispatchBuilder::FIST(OpcodeArgs, bool Truncate) {
 
   Data = _F80CVTInt(Size, Data, Truncate);
 
-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, Data, Size, OpSize::i8Bit);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, Data, Size, OpSize::i8Bit);
 
   if ((Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) != 0) {
     _PopStackDestroy();
@@ -206,10 +204,10 @@ void OpDispatchBuilder::FADD(OpcodeArgs, IR::OpSize Width, bool Integer, OpDispa
   // We have one memory argument
   Ref Arg {};
   if (Integer) {
-    Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
     Arg = _F80CVTToInt(Arg, Width);
   } else {
-    Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     Arg = _F80CVTTo(Arg, Width);
   }
 
@@ -236,10 +234,10 @@ void OpDispatchBuilder::FMUL(OpcodeArgs, IR::OpSize Width, bool Integer, OpDispa
   // We have one memory argument
   Ref arg {};
   if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
     arg = _F80CVTToInt(arg, Width);
   } else {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     arg = _F80CVTTo(arg, Width);
   }
 
@@ -273,10 +271,10 @@ void OpDispatchBuilder::FDIV(OpcodeArgs, IR::OpSize Width, bool Integer, bool Re
   // We have one memory argument
   Ref arg {};
   if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
     arg = _F80CVTToInt(arg, Width);
   } else {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     arg = _F80CVTTo(arg, Width);
   }
 
@@ -314,10 +312,10 @@ void OpDispatchBuilder::FSUB(OpcodeArgs, IR::OpSize Width, bool Integer, bool Re
   // We have one memory argument
   Ref Arg {};
   if (Integer) {
-    Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
     Arg = _F80CVTToInt(Arg, Width);
   } else {
-    Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     Arg = _F80CVTTo(Arg, Width);
   }
 
@@ -340,7 +338,7 @@ Ref OpDispatchBuilder::GetX87FTW_Helper() {
   // bytes, we use the well-known bit twiddling algorithm:
   //
   // https://graphics.stanford.edu/~seander/bithacks.html#InterleaveBMN
-  Ref X = _LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+  Ref X = _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
   X = _Orlshl(OpSize::i32Bit, X, X, 4);
   X = _And(OpSize::i32Bit, X, Constant(0x0f0f0f0f));
   X = _Orlshl(OpSize::i32Bit, X, X, 2);
@@ -381,41 +379,41 @@ void OpDispatchBuilder::X87FNSTENV(OpcodeArgs) {
   _SyncStackToSlow();
 
   const auto Size = OpSizeFromSrc(Op);
-  Ref Mem = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
+  Ref Mem = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});
   Mem = AppendSegmentOffset(Mem, Op->Flags);
 
   {
-    auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-    _StoreMem(GPRClass, Size, Mem, FCW, Size);
+    auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+    _StoreMemGPR(Size, Mem, FCW, Size);
   }
 
-  { _StoreMem(GPRClass, Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1); }
+  { _StoreMemGPR(Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1); }
 
   auto ZeroConst = Constant(0);
 
   {
     // FTW
-    _StoreMem(GPRClass, Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1);
   }
 
   {
     // Instruction Offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MemOffsetType::SXTX, 1);
   }
 
   {
     // Instruction CS selector (+ Opcode)
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MemOffsetType::SXTX, 1);
   }
 
   {
     // Data pointer offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MemOffsetType::SXTX, 1);
   }
 
   {
     // Data pointer selector
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MemOffsetType::SXTX, 1);
   }
 }
 
@@ -441,20 +439,20 @@ void OpDispatchBuilder::X87LDENV(OpcodeArgs) {
   _StackForceSlow();
 
   const auto Size = OpSizeFromSrc(Op);
-  Ref Mem = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, {.LoadData = false});
+  Ref Mem = LoadSourceGPR(Op, Op->Src[0], Op->Flags, {.LoadData = false});
   Mem = AppendSegmentOffset(Mem, Op->Flags);
 
-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 
   Ref MemLocation = Add(OpSize::i64Bit, Mem, IR::OpSizeToSize(Size) * 1);
-  auto NewFSW = _LoadMem(GPRClass, Size, MemLocation, Size);
+  auto NewFSW = _LoadMemGPR(Size, MemLocation, Size);
   ReconstructX87StateFromFSW_Helper(NewFSW);
 
   {
     // FTW
     Ref MemLocation = Add(OpSize::i64Bit, Mem, IR::OpSizeToSize(Size) * 2);
-    SetX87FTW(_LoadMem(GPRClass, Size, MemLocation, Size));
+    SetX87FTW(_LoadMemGPR(Size, MemLocation, Size));
   }
 }
 
@@ -483,61 +481,61 @@ void OpDispatchBuilder::X87FNSAVE(OpcodeArgs) {
   Ref Mem = MakeSegmentAddress(Op, Op->Dest);
   Ref Top = GetX87Top();
   {
-    auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-    _StoreMem(GPRClass, Size, Mem, FCW, Size);
+    auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+    _StoreMemGPR(Size, Mem, FCW, Size);
   }
 
-  { _StoreMem(GPRClass, Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1); }
+  { _StoreMemGPR(Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1); }
 
   auto ZeroConst = Constant(0);
 
   {
     // FTW
-    _StoreMem(GPRClass, Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1);
   }
 
   {
     // Instruction Offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MemOffsetType::SXTX, 1);
   }
 
   {
     // Instruction CS selector (+ Opcode)
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MemOffsetType::SXTX, 1);
   }
 
   {
     // Data pointer offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MemOffsetType::SXTX, 1);
   }
 
   {
     // Data pointer selector
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MemOffsetType::SXTX, 1);
   }
 
   auto SevenConst = Constant(7);
   const auto LoadSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
   for (int i = 0; i < 7; ++i) {
-    Ref data = _LoadContextIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+    Ref data = _LoadContextFPRIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
     if (ReducedPrecisionMode) {
       data = _F80CVTTo(data, OpSize::i64Bit);
     }
-    _StoreMem(FPRClass, OpSize::i128Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+    _StoreMemFPR(OpSize::i128Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
     Top = _And(OpSize::i32Bit, Add(OpSize::i32Bit, Top, 1), SevenConst);
   }
 
   // The final st(7) needs a bit of special handling here
-  Ref data = _LoadContextIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+  Ref data = _LoadContextFPRIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
   if (ReducedPrecisionMode) {
     data = _F80CVTTo(data, OpSize::i64Bit);
   }
   // ST7 broken in to two parts
   // Lower 64bits [63:0]
   // upper 16 bits [79:64]
-  _StoreMem(FPRClass, OpSize::i64Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  _StoreMemFPR(OpSize::i64Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
   auto topBytes = _VDupElement(OpSize::i128Bit, OpSize::i16Bit, data, 4);
-  _StoreMem(FPRClass, OpSize::i16Bit, topBytes, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10) + 8), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  _StoreMemFPR(OpSize::i16Bit, topBytes, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10) + 8), OpSize::i8Bit, MemOffsetType::SXTX, 1);
 
   // reset to default
   FNINIT(Op);
@@ -548,8 +546,8 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
   const auto Size = OpSizeFromSrc(Op);
   Ref Mem = MakeSegmentAddress(Op, Op->Src[0]);
 
-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
   if (ReducedPrecisionMode) {
     // ignore the rounding precision, we're always 64-bit in F64.
     // extract rounding mode
@@ -561,11 +559,11 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
     _SetRoundingMode(roundingMode, false, roundingMode);
   }
 
-  auto NewFSW = _LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1);
+  auto NewFSW = _LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1);
   Ref Top = ReconstructX87StateFromFSW_Helper(NewFSW);
   {
     // FTW
-    SetX87FTW(_LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1));
+    SetX87FTW(_LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1));
   }
 
   auto SevenConst = Constant(7);
@@ -574,14 +572,14 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
   Ref Mask = _VLoadTwoGPRs(low, high);
   const auto StoreSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
   for (int i = 0; i < 7; ++i) {
-    Ref Reg = _LoadMem(FPRClass, OpSize::i128Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+    Ref Reg = _LoadMemFPR(OpSize::i128Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
     // Mask off the top bits
     Reg = _VAnd(OpSize::i128Bit, OpSize::i128Bit, Reg, Mask);
     if (ReducedPrecisionMode) {
       // Convert to double precision
       Reg = _F80CVT(OpSize::i64Bit, Reg);
     }
-    _StoreContextIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+    _StoreContextFPRIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
 
     Top = _And(OpSize::i32Bit, Add(OpSize::i32Bit, Top, 1), SevenConst);
   }
@@ -590,19 +588,19 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
   // ST7 broken in to two parts
   // Lower 64bits [63:0]
   // upper 16 bits [79:64]
-  Ref Reg = _LoadMem(FPRClass, OpSize::i64Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
-  Ref RegHigh = _LoadMem(FPRClass, OpSize::i16Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7) + 8), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  Ref Reg = _LoadMemFPR(OpSize::i64Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
+  Ref RegHigh = _LoadMemFPR(OpSize::i16Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7) + 8), OpSize::i8Bit, MemOffsetType::SXTX, 1);
   Reg = _VInsElement(OpSize::i128Bit, OpSize::i16Bit, 4, 0, Reg, RegHigh);
   if (ReducedPrecisionMode) {
     Reg = _F80CVT(OpSize::i64Bit, Reg); // Convert to double precision
   }
-  _StoreContextIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+  _StoreContextFPRIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
 }
 
 // Load / Store Control Word
 void OpDispatchBuilder::X87FSTCW(OpcodeArgs) {
-  auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-  StoreResult(GPRClass, Op, FCW, OpSize::iInvalid);
+  auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+  StoreResultGPR(Op, FCW);
 }
 
 void OpDispatchBuilder::X87FLDCW(OpcodeArgs) {
@@ -610,8 +608,8 @@ void OpDispatchBuilder::X87FLDCW(OpcodeArgs) {
   // to switch for now to slow mode whenever these are manually changed.
   // Remove the next line and try DF_04.asm in fast path.
   _StackForceSlow();
-  Ref NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  Ref NewFCW = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 }
 
 void OpDispatchBuilder::FXCH(OpcodeArgs) {
@@ -647,10 +645,10 @@ void OpDispatchBuilder::FCOMI(OpcodeArgs, IR::OpSize Width, bool Integer, OpDisp
     if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
       // Memory arg
       if (Integer) {
-        arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+        arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
         b = _F80CVTToInt(arg, Width);
       } else {
-        arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+        arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
         b = _F80CVTTo(arg, Width);
       }
     } else {
@@ -766,7 +764,7 @@ Ref OpDispatchBuilder::ReconstructFSW_Helper(Ref T) {
 void OpDispatchBuilder::X87FNSTSW(OpcodeArgs) {
   Ref TopValue = _SyncStackToSlow();
   Ref StatusWord = ReconstructFSW_Helper(TopValue);
-  StoreResult(GPRClass, Op, StatusWord, OpSize::iInvalid);
+  StoreResultGPR(Op, StatusWord);
 }
 
 void OpDispatchBuilder::FNCLEX(OpcodeArgs) {
@@ -785,12 +783,12 @@ void OpDispatchBuilder::FNINIT(OpcodeArgs) {
 
   // Init FCW to 0x037F
   auto NewFCW = Constant(0x037F);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 
   // Set top to zero
   SetX87Top(Zero);
   // Tags all get marked as invalid
-  _StoreContext(OpSize::i8Bit, GPRClass, Zero, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+  _StoreContextGPR(OpSize::i8Bit, Zero, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
 
   // Reinits the simulated stack
   _InitStack();
@@ -863,7 +861,7 @@ void OpDispatchBuilder::X87FXAM(OpcodeArgs) {
   auto TopValid = _StackValidTag(0);
 
   // In the case of top being invalid then C3:C2:C0 is 0b101
-  auto C3 = Select01(OpSize::i32Bit, CondClassType {COND_NEQ}, TopValid, Constant(1));
+  auto C3 = Select01(OpSize::i32Bit, CondClass::NEQ, TopValid, Constant(1));
 
   auto C2 = TopValid;
   auto C0 = C3; // Mirror C3 until something other than zero is supported
@@ -878,8 +876,8 @@ void OpDispatchBuilder::X87FXTRACT(OpcodeArgs) {
   _PopStackDestroy();
   auto Exp = _F80XTRACT_EXP(Top);
   auto Sig = _F80XTRACT_SIG(Top);
-  _PushStack(Exp, Exp, OpSize::f80Bit, true);
-  _PushStack(Sig, Sig, OpSize::f80Bit, true);
+  _PushStack(Exp, Invalid(), OpSize::iInvalid);
+  _PushStack(Sig, Invalid(), OpSize::iInvalid);
 }
 
 } // namespace FEXCore::IR

FEXCore/Source/Interface/Core/OpcodeDispatcher/X87F64.cpp

@@ -29,38 +29,37 @@ void OpDispatchBuilder::X87LDENVF64(OpcodeArgs) {
   const auto Size = OpSizeFromSrc(Op);
   Ref Mem = MakeSegmentAddress(Op, Op->Src[0]);
 
-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
   // ignore the rounding precision, we're always 64-bit in F64.
   // extract rounding mode
   Ref roundingMode = _Bfe(OpSize::i32Bit, 3, 10, NewFCW);
   _SetRoundingMode(roundingMode, false, roundingMode);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 
-  auto NewFSW = _LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size)), Size, MEM_OFFSET_SXTX, 1);
+  auto NewFSW = _LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size)), Size, MemOffsetType::SXTX, 1);
   ReconstructX87StateFromFSW_Helper(NewFSW);
 
   {
     // FTW
-    SetX87FTW(_LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1));
+    SetX87FTW(_LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1));
   }
 }
 
 void OpDispatchBuilder::X87FLDCWF64(OpcodeArgs) {
   _StackForceSlow();
 
-  Ref NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+  Ref NewFCW = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
   // ignore the rounding precision, we're always 64-bit in F64.
   // extract rounding mode
   Ref roundingMode = _Bfe(OpSize::i32Bit, 3, 10, NewFCW);
   _SetRoundingMode(roundingMode, false, roundingMode);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 }
 
 // F64 ops
 // Float load op with memory operand
 void OpDispatchBuilder::FLDF64(OpcodeArgs, IR::OpSize Width) {
-  const auto ReadWidth = (Width == OpSize::f80Bit) ? OpSize::i128Bit : Width;
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], Width, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], Width, Op->Flags);
   // Convert to 64bit float
   Ref ConvertedData = Data;
   if (Width == OpSize::i32Bit) {
@@ -68,39 +67,39 @@ void OpDispatchBuilder::FLDF64(OpcodeArgs, IR::OpSize Width) {
   } else if (Width == OpSize::f80Bit) {
     ConvertedData = _F80CVT(OpSize::i64Bit, Data);
   }
-  _PushStack(ConvertedData, Data, ReadWidth, true);
+  _PushStack(ConvertedData, Data, Width);
 }
 
 void OpDispatchBuilder::FBLDF64(OpcodeArgs) {
   // Read from memory
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
   Ref ConvertedData = _F80BCDLoad(Data);
   ConvertedData = _F80CVT(OpSize::i64Bit, ConvertedData);
-  _PushStack(ConvertedData, Data, OpSize::i64Bit, true);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }
 
 void OpDispatchBuilder::FBSTPF64(OpcodeArgs) {
   Ref converted = _F80CVTTo(_ReadStackValue(0), OpSize::i64Bit);
   converted = _F80BCDStore(converted);
-  StoreResult_WithOpSize(FPRClass, Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
+  StoreResultFPR_WithOpSize(Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
   _PopStackDestroy();
 }
 
 void OpDispatchBuilder::FLDF64_Const(OpcodeArgs, uint64_t Num) {
   auto Data = _VCastFromGPR(OpSize::i64Bit, OpSize::i64Bit, Constant(Num));
-  _PushStack(Data, Data, OpSize::i64Bit, true);
+  _PushStack(Data, Data, OpSize::i64Bit);
 }
 
 void OpDispatchBuilder::FILDF64(OpcodeArgs) {
   const auto ReadWidth = OpSizeFromSrc(Op);
 
   // Read from memory
-  Ref Data = LoadSource_WithOpSize(GPRClass, Op, Op->Src[0], ReadWidth, Op->Flags);
+  Ref Data = LoadSourceGPR_WithOpSize(Op, Op->Src[0], ReadWidth, Op->Flags);
   if (ReadWidth == OpSize::i16Bit) {
     Data = _Sbfe(OpSize::i64Bit, IR::OpSizeAsBits(ReadWidth), 0, Data);
   }
   auto ConvertedData = _Float_FromGPR_S(OpSize::i64Bit, ReadWidth == OpSize::i32Bit ? OpSize::i32Bit : OpSize::i64Bit, Data);
-  _PushStack(ConvertedData, Data, ReadWidth, false);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }
 
 void OpDispatchBuilder::FISTF64(OpcodeArgs, bool Truncate) {
@@ -112,7 +111,7 @@ void OpDispatchBuilder::FISTF64(OpcodeArgs, bool Truncate) {
   } else {
     data = _Float_ToGPR_S(Size == OpSize::i32Bit ? OpSize::i32Bit : OpSize::i64Bit, OpSize::i64Bit, data);
   }
-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, data, Size, OpSize::i8Bit);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, data, Size, OpSize::i8Bit);
 
   if ((Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) != 0) {
     _PopStackDestroy();
@@ -138,16 +137,16 @@ void OpDispatchBuilder::FADDF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpDi
   Ref arg {};
 
   if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
     if (Width == OpSize::i16Bit) {
       arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
     }
     arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
   } else if (Width == OpSize::i32Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
   } else if (Width == OpSize::i64Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   } else {
     FEX_UNREACHABLE;
   }
@@ -176,16 +175,16 @@ void OpDispatchBuilder::FMULF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpDi
   Ref arg {};
 
   if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
     if (Width == OpSize::i16Bit) {
       arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
     }
     arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
   } else if (Width == OpSize::i32Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
   } else if (Width == OpSize::i64Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
   } else {
     FEX_UNREACHABLE;
   }
@@ -228,16 +227,16 @@ void OpDispatchBuilder::FDIVF64(OpcodeArgs, IR::OpSize Width, bool Integer, bool
 
   if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
     if (Integer) {
-      Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
       if (Width == OpSize::i16Bit) {
         Arg = _Sbfe(OpSize::i64Bit, 16, 0, Arg);
       }
       Arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, Arg);
     } else if (Width == OpSize::i32Bit) {
-      Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
       Arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, Arg);
     } else if (Width == OpSize::i64Bit) {
-      Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     }
   } else {
     FEX_UNREACHABLE;
@@ -285,16 +284,16 @@ void OpDispatchBuilder::FSUBF64(OpcodeArgs, IR::OpSize Width, bool Integer, bool
 
   if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
     if (Integer) {
-      arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
       if (Width == OpSize::i16Bit) {
         arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
       }
       arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
     } else if (Width == OpSize::i32Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
       arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
     } else if (Width == OpSize::i64Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     }
   } else {
     FEX_UNREACHABLE;
@@ -332,16 +331,16 @@ void OpDispatchBuilder::FCOMIF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpD
   } else if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
     // Memory arg
     if (Integer) {
-      arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
       if (Width == OpSize::i16Bit) {
         arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
       }
       b = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
     } else if (Width == OpSize::i32Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
       b = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
     } else if (Width == OpSize::i64Bit) {
-      b = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      b = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
     }
   } else {
     FEX_UNREACHABLE;
@@ -393,11 +392,11 @@ void OpDispatchBuilder::X87FXTRACTF64(OpcodeArgs) {
   SaveNZCV();
   _TestNZ(OpSize::i64Bit, Gpr, Constant(0x7fff'ffff'ffff'ffffUL));
 
-  Ref Sig = _NZCVSelectV(OpSize::i64Bit, {COND_EQ}, SigZV, SigNZV);
-  Ref Exp = _NZCVSelectV(OpSize::i64Bit, {COND_EQ}, ExpZV, ExpNZV);
+  Ref Sig = _NZCVSelectV(OpSize::i64Bit, CondClass::EQ, SigZV, SigNZV);
+  Ref Exp = _NZCVSelectV(OpSize::i64Bit, CondClass::EQ, ExpZV, ExpNZV);
 
   _PopStackDestroy();
-  _PushStack(Exp, Exp, OpSize::i64Bit, true);
-  _PushStack(Sig, Sig, OpSize::i64Bit, true);
+  _PushStack(Exp, Invalid(), OpSize::iInvalid);
+  _PushStack(Sig, Invalid(), OpSize::iInvalid);
 }
 } // namespace FEXCore::IR

FEXCore/Source/Interface/Core/X86HelperGen.cpp

@@ -1,87 +0,0 @@
-// SPDX-License-Identifier: MIT
-/*
-$info$
-tags: glue|x86-guest-code
-desc: Guest-side assembly helpers used by the backends
-$end_info$
-*/
-
-#include "Interface/Core/X86HelperGen.h"
-#include "FEXCore/Utils/AllocatorHooks.h"
-
-#include <FEXCore/Config/Config.h>
-#include <FEXCore/Utils/Allocator.h>
-#include <FEXHeaderUtils/Syscalls.h>
-
-#include <cstdint>
-#include <cstring>
-
-namespace FEXCore {
-constexpr size_t CODE_SIZE = 0x1000;
-
-X86GeneratedCode::X86GeneratedCode() {
-#ifdef _WIN32
-  // No need to allocate anything in this config.
-#else
-
-  // Allocate a page for our emulated guest
-  CodePtr = AllocateGuestCodeSpace(CODE_SIZE);
-
-  constexpr std::array<uint8_t, 2> SignalReturnCode = {
-    0x0F, 0x37, // CALLBACKRET FEX Instruction
-  };
-
-  CallbackReturn = reinterpret_cast<uint64_t>(CodePtr);
-
-  memcpy(reinterpret_cast<void*>(CallbackReturn), SignalReturnCode.data(), SignalReturnCode.size());
-
-  mprotect(CodePtr, CODE_SIZE, PROT_READ);
-#endif
-}
-
-X86GeneratedCode::~X86GeneratedCode() {
-#ifndef _WIN32
-  FEXCore::Allocator::VirtualFree(CodePtr, CODE_SIZE);
-#endif
-}
-
-void* X86GeneratedCode::AllocateGuestCodeSpace(size_t Size) {
-#ifndef _WIN32
-  FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
-
-  if (Is64BitMode()) {
-    // 64bit mode can have its sigret handler anywhere
-    return FEXCore::Allocator::VirtualAlloc(Size);
-  }
-
-  // First 64bit page
-  constexpr uintptr_t LOCATION_MAX = 0x1'0000'0000;
-
-  // 32bit mode
-  // We need to have the sigret handler in the lower 32bits of memory space
-  // Scan top down and try to allocate a location
-  for (size_t Location = 0xFFFF'E000; Location != 0x0; Location -= 0x1000) {
-    void* Ptr = ::mmap(reinterpret_cast<void*>(Location), Size, PROT_READ | PROT_WRITE, MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
-
-    if (Ptr != MAP_FAILED && reinterpret_cast<uintptr_t>(Ptr) >= LOCATION_MAX) {
-      // Failed to map in the lower 32bits
-      // Try again
-      // Can happen in the case that host kernel ignores MAP_FIXED_NOREPLACE
-      ::munmap(Ptr, Size);
-      continue;
-    }
-
-    if (Ptr != MAP_FAILED) {
-      return Ptr;
-    }
-  }
-
-  // Can't do anything about this
-  // Here's hoping the application doesn't use signals
-  return MAP_FAILED;
-#else
-  return nullptr;
-#endif
-}
-
-} // namespace FEXCore

FEXCore/Source/Interface/Core/X86HelperGen.h

@@ -1,25 +0,0 @@
-// SPDX-License-Identifier: MIT
-/*
-$info$
-tags: glue|x86-guest-code
-$end_info$
-*/
-
-#pragma once
-
-#include <stddef.h>
-#include <stdint.h>
-
-namespace FEXCore {
-class X86GeneratedCode final {
-public:
-  X86GeneratedCode();
-  ~X86GeneratedCode();
-
-  uint64_t CallbackReturn {};
-
-private:
-  void* CodePtr {};
-  void* AllocateGuestCodeSpace(size_t Size);
-};
-} // namespace FEXCore

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

@@ -100,10 +100,11 @@ constexpr std::array<X86InstInfo, MAX_SECOND_TABLE_SIZE> SecondBaseOps = []() co
     {0x34, 1, X86InstInfo{"SYSENTER",   TYPE_INST, FLAGS_NO_OVERLAY,                                                                             0}},
     {0x35, 1, X86InstInfo{"SYSEXIT",    TYPE_INST, FLAGS_NO_OVERLAY,                                                                             0}},
     {0x36, 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
+    {0x37, 1, X86InstInfo{"GETSEC",     TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
     {0x38, 1, X86InstInfo{"",           TYPE_0F38_TABLE, FLAGS_NO_OVERLAY,                                                                       0}},
     {0x39, 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
     {0x3A, 1, X86InstInfo{"",           TYPE_0F3A_TABLE, FLAGS_NO_OVERLAY,                                                                       0}},
-    {0x3B, 4, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
+    {0x3B, 3, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
 
     {0x40, 1, X86InstInfo{"CMOVO",      TYPE_INST, FLAGS_MODRM | FLAGS_NO_OVERLAY,                                                               0}},
     {0x41, 1, X86InstInfo{"CMOVNO",     TYPE_INST, FLAGS_MODRM | FLAGS_NO_OVERLAY,                                                               0}},
@@ -299,7 +300,7 @@ constexpr std::array<X86InstInfo, MAX_SECOND_TABLE_SIZE> SecondBaseOps = []() co
     // FEX reserved instructions
     // Unused x86 encoding instruction.
 
-    {0x37, 1, X86InstInfo{"CALLBACKRET",  TYPE_INST, FLAGS_BLOCK_END | FLAGS_NO_OVERLAY | FLAGS_SETS_RIP,                                                                          0}},
+    {0x3E, 1, X86InstInfo{"CALLBACKRET",  TYPE_INST, FLAGS_BLOCK_END | FLAGS_NO_OVERLAY | FLAGS_SETS_RIP,                                                                          0}},
 
     // This was originally used by VIA to jump to its alternative instruction set. Used for OP_THUNK
     {0x3F, 1, X86InstInfo{"ALTINST",      TYPE_INST, FLAGS_BLOCK_END | FLAGS_NO_OVERLAY | FLAGS_SETS_RIP,                                                            0}},
@@ -442,7 +443,7 @@ constexpr std::array<X86InstInfo, MAX_REPNE_MOD_TABLE_SIZE> RepNEModOps = []() c
     {0x70, 1, X86InstInfo{"PSHUFLW",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                   1}},
     {0x71, 3, X86InstInfo{"",          TYPE_COPY_OTHER, FLAGS_NONE,                                                      0}},
     {0x74, 4, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                                         0}},
-    {0x78, 1, X86InstInfo{"INSERTQ",   TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY | FLAGS_XMM_FLAGS,2}},
+    {0x78, 1, X86InstInfo{"INSERTQ",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY | FLAGS_XMM_FLAGS,2}},
     {0x79, 1, X86InstInfo{"INSERTQ",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY | FLAGS_XMM_FLAGS, 0}},
     {0x7A, 2, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                                         0}},
     {0x7C, 1, X86InstInfo{"HADDPS",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                   0}},

FEXCore/Source/Interface/Core/X86Tables/X86Tables.h

@@ -7,6 +7,7 @@ $end_info$
 
 #pragma once
 
+#include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/LogManager.h>
 
 #include <array>
@@ -559,21 +560,6 @@ constexpr static inline void GenerateTableWithCopy(X86InstInfo *FinalTable, X86T
   }
 };
 
-template<typename OpcodeType>
-static inline void LateInitCopyTable(X86InstInfo *FinalTable, X86TablesInfoStruct<OpcodeType> const *OtherLocal, size_t OtherTableSize) {
-  for (size_t j = 0; j < OtherTableSize; ++j) {
-    X86TablesInfoStruct<OpcodeType> const &OtherOp = OtherLocal[j];
-    auto OtherOpNum = OtherOp.first;
-    X86InstInfo const &OtherInfo = OtherOp.Info;
-    for (uint32_t i = 0; i < OtherOp.second; ++i) {
-      X86InstInfo &FinalOp = FinalTable[OtherOpNum + i];
-      if (FinalOp.Type == TYPE_COPY_OTHER) {
-        FinalOp = OtherInfo;
-      }
-    }
-  }
-}
-
 template<typename OpcodeType>
 constexpr static inline void GenerateX87Table(X86InstInfo *FinalTable, X86TablesInfoStruct<OpcodeType> const *LocalTable, size_t TableSize) {
   for (size_t j = 0; j < TableSize; ++j) {
@@ -604,12 +590,6 @@ constexpr static inline void GenerateX87Table(X86InstInfo *FinalTable, X86Tables
   }
 };
 
-}
+FEX_DEFINE_ENUM_FMT_PASSTHROUGH(FEXCore::X86Tables::DecodedOperand::OpType);
 
-template <>
-struct fmt::formatter<FEXCore::X86Tables::DecodedOperand::OpType> : formatter<uint32_t> {
-  template <typename FormatContext>
-  auto format(FEXCore::X86Tables::DecodedOperand::OpType type, FormatContext& ctx) const {
-    return fmt::formatter<uint32_t>::format(static_cast<uint32_t>(type), ctx);
-  }
-};
+} // namespace FEXCore::X86Tables

FEXCore/Source/Interface/IR/IR.h

@@ -20,7 +20,6 @@
 namespace FEXCore::IR {
 
 class OrderedNode;
-class RegisterAllocationPass;
 
 /**
  * @brief The IROp_Header is an dynamically sized array
@@ -61,24 +60,7 @@ struct NodeID final {
     Value = 0;
   }
 
-  [[nodiscard]] friend constexpr bool operator==(NodeID, NodeID) noexcept = default;
-
-  [[nodiscard]]
-  friend constexpr bool operator<(NodeID lhs, NodeID rhs) noexcept {
-    return lhs.Value < rhs.Value;
-  }
-  [[nodiscard]]
-  friend constexpr bool operator>(NodeID lhs, NodeID rhs) noexcept {
-    return operator<(rhs, lhs);
-  }
-  [[nodiscard]]
-  friend constexpr bool operator<=(NodeID lhs, NodeID rhs) noexcept {
-    return !operator>(lhs, rhs);
-  }
-  [[nodiscard]]
-  friend constexpr bool operator>=(NodeID lhs, NodeID rhs) noexcept {
-    return !operator<(lhs, rhs);
-  }
+  [[nodiscard]] constexpr auto operator<=>(const NodeID&) const noexcept = default;
 
   friend std::ostream& operator<<(std::ostream& out, NodeID ID) {
     out << ID.Value;
@@ -431,94 +413,6 @@ static_assert(sizeof(OrderedNode) == (sizeof(OrderedNodeHeader) + 2 * sizeof(uin
 //  };
 using Ref = OrderedNode*;
 
-struct FEX_PACKED RegisterClassType final {
-  using value_type = uint32_t;
-
-  value_type Val;
-  [[nodiscard]] constexpr operator value_type() const {
-    return Val;
-  }
-  [[nodiscard]]
-  friend constexpr bool operator==(const RegisterClassType&, const RegisterClassType&) = default;
-};
-
-struct FEX_PACKED CondClassType final {
-  uint8_t Val;
-  [[nodiscard]] constexpr operator uint8_t() const {
-    return Val;
-  }
-  [[nodiscard]]
-  friend constexpr bool operator==(const CondClassType&, const CondClassType&) = default;
-};
-
-struct FEX_PACKED MemOffsetType final {
-  uint8_t Val;
-  [[nodiscard]] constexpr operator uint8_t() const {
-    return Val;
-  }
-  [[nodiscard]]
-  friend constexpr bool operator==(const MemOffsetType&, const MemOffsetType&) = default;
-};
-
-struct FEX_PACKED TypeDefinition final {
-  uint16_t Val;
-
-  [[nodiscard]] constexpr operator uint16_t() const {
-    return Val;
-  }
-
-  [[nodiscard]]
-  static constexpr TypeDefinition Create(uint8_t Bytes) {
-    TypeDefinition Type {};
-    Type.Val = Bytes << 8;
-    return Type;
-  }
-
-  [[nodiscard]]
-  static constexpr TypeDefinition Create(uint8_t Bytes, uint8_t Elements) {
-    TypeDefinition Type {};
-    Type.Val = (Bytes << 8) | (Elements & 255);
-    return Type;
-  }
-
-  [[nodiscard]]
-  constexpr uint8_t Bytes() const {
-    return Val >> 8;
-  }
-
-  [[nodiscard]]
-  constexpr uint8_t Elements() const {
-    return Val & 255;
-  }
-
-  [[nodiscard]]
-  friend constexpr bool operator==(const TypeDefinition&, const TypeDefinition&) = default;
-};
-
-static_assert(std::is_trivially_copyable_v<TypeDefinition>);
-
-struct FEX_PACKED FenceType final {
-  using value_type = uint8_t;
-
-  value_type Val;
-  [[nodiscard]] constexpr operator value_type() const {
-    return Val;
-  }
-  [[nodiscard]]
-  friend constexpr bool operator==(const FenceType&, const FenceType&) = default;
-};
-
-struct FEX_PACKED RoundType final {
-  uint8_t Val;
-  [[nodiscard]] constexpr operator uint8_t() const {
-    return Val;
-  }
-  [[nodiscard]]
-  friend constexpr bool operator==(const RoundType&, const RoundType&) = default;
-};
-
-class NodeIterator;
-
 /* This iterator can be used to step though nodes.
  * Due to how our IR is laid out, this can be used to either step
  * though the CodeBlocks or though the code within a single block.
@@ -782,6 +676,15 @@ inline NodeID NodeWrapperBase<Type>::ID() const {
 bool IsBlockExit(FEXCore::IR::IROps Op);
 
 void Dump(fextl::stringstream* out, const IRListView* IR);
+
+constexpr auto format_as(FEXCore::IR::NodeID ID) {
+  return ID.Value;
+}
+
+FEX_DEFINE_ENUM_FMT_PASSTHROUGH(FEXCore::IR::FenceType)
+FEX_DEFINE_ENUM_FMT_PASSTHROUGH(FEXCore::IR::MemOffsetType)
+FEX_DEFINE_ENUM_FMT_PASSTHROUGH(FEXCore::IR::OpSize)
+FEX_DEFINE_ENUM_FMT_PASSTHROUGH(FEXCore::IR::RegClass)
 } // namespace FEXCore::IR
 
 template<>
@@ -790,45 +693,3 @@ struct std::hash<FEXCore::IR::NodeID> {
     return std::hash<FEXCore::IR::NodeID::value_type> {}(ID.Value);
   }
 };
-
-template<>
-struct fmt::formatter<FEXCore::IR::NodeID> : fmt::formatter<FEXCore::IR::NodeID::value_type> {
-  using Base = fmt::formatter<FEXCore::IR::NodeID::value_type>;
-
-  // Pass-through the underlying value, so IDs can
-  // be formatted like any integral value.
-  template<typename FormatContext>
-  auto format(const FEXCore::IR::NodeID& ID, FormatContext& ctx) const {
-    return Base::format(ID.Value, ctx);
-  }
-};
-
-template<>
-struct fmt::formatter<FEXCore::IR::RegisterClassType> : fmt::formatter<FEXCore::IR::RegisterClassType::value_type> {
-  using Base = fmt::formatter<FEXCore::IR::RegisterClassType::value_type>;
-
-  template<typename FormatContext>
-  auto format(const FEXCore::IR::RegisterClassType& Class, FormatContext& ctx) const {
-    return Base::format(Class.Val, ctx);
-  }
-};
-
-template<>
-struct fmt::formatter<FEXCore::IR::FenceType> : fmt::formatter<FEXCore::IR::FenceType::value_type> {
-  using Base = fmt::formatter<FEXCore::IR::FenceType::value_type>;
-
-  template<typename FormatContext>
-  auto format(const FEXCore::IR::FenceType& Fence, FormatContext& ctx) const {
-    return Base::format(Fence.Val, ctx);
-  }
-};
-
-template<>
-struct fmt::formatter<FEXCore::IR::OpSize> : fmt::formatter<std::underlying_type_t<FEXCore::IR::OpSize>> {
-  using Base = fmt::formatter<std::underlying_type_t<FEXCore::IR::OpSize>>;
-
-  template<typename FormatContext>
-  auto format(const FEXCore::IR::OpSize& OpSize, FormatContext& ctx) const {
-    return Base::format(FEXCore::ToUnderlying(OpSize), ctx);
-  }
-};

FEXCore/Source/Interface/IR/IR.json

@@ -52,80 +52,68 @@
     "  * These are validations that can't be automatically inferred and need to be hand-written",
     ""
   ],
+  "Enums": {
+    "class CondClass : uint8_t": [
+      "EQ    = 0,",
+      "NEQ   = 1,",
+      "UGE   = 2,",
+      "ULT   = 3,",
+      "MI    = 4,",
+      "PL    = 5,",
+      "VS    = 6,",
+      "VC    = 7,",
+      "UGT   = 8,",
+      "ULE   = 9,",
+      "SGE   = 10,",
+      "SLT   = 11,",
+      "SGT   = 12,",
+      "SLE   = 13,",
+      "TSTZ  = 14, /* bit test zero */",
+      "TSTNZ = 15, /* bit test nonzero */",
+      "",
+      "FLU   = 16, /* float less or unordered */",
+      "FGE   = 17, /* float greater or equal */",
+      "FLEU  = 18, /* float less or equal or unordered */",
+      "FGT   = 19, /* float greater */",
+      "FU    = 20, /* float unordered */",
+      "FNU   = 21, /* float not unordered */",
+      "",
+      "AL    = 32, /* always */"
+    ],
+    "class FenceType : uint8_t": [
+      "Load      = 0,",
+      "Store     = 1,",
+      "LoadStore = 2,",
+      "Inst      = 3,"
+    ],
+    "class MemOffsetType : uint8_t": [
+      "SXTX = 0,",
+      "UXTW = 1,",
+      "SXTW = 2,"
+    ],
+    "class RegClass : uint32_t": [
+      "Invalid  = 0,",
+      "GPR      = 1,",
+      "GPRFixed = 2,",
+      "FPR      = 3,",
+      "FPRFixed = 4,",
+      "Complex  = 5,"
+    ],
+    "class RoundMode : uint8_t": [
+      "Nearest     = 0,",
+      "NegInfinity = 1,",
+      "PosInfinity = 2,",
+      "TowardsZero = 3, /* Truncate */",
+      "Host        = 4,"
+    ]
+  },
   "Defines": [
-    "constexpr uint8_t COND_EQ  = 0",
-    "constexpr uint8_t COND_NEQ = 1",
-    "constexpr uint8_t COND_UGE  = 2",
-    "constexpr uint8_t COND_ULT  = 3",
-    "constexpr uint8_t COND_MI  = 4",
-    "constexpr uint8_t COND_PL  = 5",
-    "constexpr uint8_t COND_VS  = 6",
-    "constexpr uint8_t COND_VC  = 7",
-    "constexpr uint8_t COND_UGT  = 8",
-    "constexpr uint8_t COND_ULE  = 9",
-    "constexpr uint8_t COND_SGE  = 10",
-    "constexpr uint8_t COND_SLT  = 11",
-    "constexpr uint8_t COND_SGT  = 12",
-    "constexpr uint8_t COND_SLE  = 13",
-    "constexpr uint8_t COND_TSTZ = 14 /* bit test zero */",
-    "constexpr uint8_t COND_TSTNZ = 15 /* bit test nonzero */",
-
-    "constexpr uint8_t COND_FLU  = 16 /* float less or unordred */",
-    "constexpr uint8_t COND_FGE  = 17 /* float greater or equal */",
-    "constexpr uint8_t COND_FLEU = 18 /* float less or equal or unordred */",
-    "constexpr uint8_t COND_FGT  = 19 /* float greater */",
-    "constexpr uint8_t COND_FU   = 20 /* float unordred */",
-    "constexpr uint8_t COND_FNU  = 21 /* float not unordred */",
-
-    "constexpr uint8_t COND_AL   = 32 /* always */",
-
-    "constexpr FEXCore::IR::RegisterClassType InvalidClass {0}",
-    "constexpr FEXCore::IR::RegisterClassType GPRClass {1}",
-    "constexpr FEXCore::IR::RegisterClassType GPRFixedClass {2}",
-    "constexpr FEXCore::IR::RegisterClassType FPRClass {3}",
-    "constexpr FEXCore::IR::RegisterClassType FPRFixedClass {4}",
-    "constexpr FEXCore::IR::RegisterClassType ComplexClass {5}",
     "constexpr uint8_t NumClasses {6}",
-    "",
-    "constexpr FEXCore::IR::TypeDefinition i8    {TypeDefinition::Create(1, 0)}",
-    "constexpr FEXCore::IR::TypeDefinition i16   {TypeDefinition::Create(2, 0)}",
-    "constexpr FEXCore::IR::TypeDefinition i32   {TypeDefinition::Create(4, 0)}",
-    "constexpr FEXCore::IR::TypeDefinition i64   {TypeDefinition::Create(8, 0)}",
-    "constexpr FEXCore::IR::TypeDefinition i128  {TypeDefinition::Create(16, 0)}",
-    "",
-    "constexpr FEXCore::IR::TypeDefinition i8v8  {TypeDefinition::Create(1, 8)}",
-    "constexpr FEXCore::IR::TypeDefinition i8v16 {TypeDefinition::Create(1, 16)}",
-    "constexpr FEXCore::IR::TypeDefinition i16v4 {TypeDefinition::Create(2, 4)}",
-    "constexpr FEXCore::IR::TypeDefinition i16v8 {TypeDefinition::Create(2, 8)}",
-    "constexpr FEXCore::IR::TypeDefinition i32v2 {TypeDefinition::Create(4, 2)}",
-    "constexpr FEXCore::IR::TypeDefinition i32v4 {TypeDefinition::Create(4, 4)}",
-    "constexpr FEXCore::IR::TypeDefinition i64v2 {TypeDefinition::Create(8, 2)}",
-    "",
+
     "constexpr uint8_t FCMP_FLAG_EQ        = 0",
     "constexpr uint8_t FCMP_FLAG_LT        = 1",
     "constexpr uint8_t FCMP_FLAG_UNORDERED = 2",
 
-    "constexpr FEXCore::IR::FenceType Fence_Load      {0}",
-    "constexpr FEXCore::IR::FenceType Fence_Store     {1}",
-    "constexpr FEXCore::IR::FenceType Fence_LoadStore {2}",
-    "constexpr FEXCore::IR::FenceType Fence_Inst      {3}",
-
-    "constexpr uint8_t ROUND_MODE_NEAREST           = 0",
-    "constexpr uint8_t ROUND_MODE_NEGATIVE_INFINITY = 1",
-    "constexpr uint8_t ROUND_MODE_POSITIVE_INFINITY = 2",
-    "constexpr uint8_t ROUND_MODE_TOWARDS_ZERO      = 3",
-    "constexpr uint8_t ROUND_MODE_FLUSH_TO_ZERO     = 1 << 2",
-
-    "constexpr FEXCore::IR::RoundType Round_Nearest {ROUND_MODE_NEAREST}",
-    "constexpr FEXCore::IR::RoundType Round_Negative_Infinity {ROUND_MODE_NEGATIVE_INFINITY}",
-    "constexpr FEXCore::IR::RoundType Round_Positive_Infinity {ROUND_MODE_POSITIVE_INFINITY}",
-    "constexpr FEXCore::IR::RoundType Round_Towards_Zero {ROUND_MODE_TOWARDS_ZERO} /* Truncate */",
-    "constexpr FEXCore::IR::RoundType Round_Host {ROUND_MODE_TOWARDS_ZERO + 1}",
-
-    "constexpr FEXCore::IR::MemOffsetType MEM_OFFSET_SXTX {0}",
-    "constexpr FEXCore::IR::MemOffsetType MEM_OFFSET_UXTW {1}",
-    "constexpr FEXCore::IR::MemOffsetType MEM_OFFSET_SXTW {2}",
-
     "struct BreakDefinition {",
     "  uint16_t ErrorRegister;",
     "  uint8_t Signal;",
@@ -148,13 +136,12 @@
     "GPR": "OrderedNode*",
     "FPR": "OrderedNode*",
     "FenceType": "FenceType",
-    "RegisterClass": "RegisterClassType",
-    "CondClass": "CondClassType",
-    "SyscallFlags": "FEXCore::IR::SyscallFlags",
+    "RegisterClass": "RegClass",
+    "CondClass": "CondClass",
     "SHA256Sum": "SHA256Sum",
     "MemOffsetType": "MemOffsetType",
     "BreakDefinition": "BreakDefinition",
-    "RoundType": "RoundType",
+    "RoundType": "RoundMode",
     "FloatCompareOp": "FloatCompareOp",
     "NamedVectorConstant": "FEXCore::IR::NamedVectorConstant",
     "IndexNamedVectorConstant": "FEXCore::IR::IndexNamedVectorConstant",
@@ -307,7 +294,7 @@
         "HasSideEffects": true,
         "RAOverride": "0"
       },
-      "CondJump SSA:$Cmp1, SSA:$Cmp2, SSA:$TrueBlock, SSA:$FalseBlock, CondClass:$Cond{{COND_NEQ}}, OpSize:$CompareSize{OpSize::iInvalid}, i1:$FromNZCV{false}": {
+      "CondJump SSA:$Cmp1, SSA:$Cmp2, SSA:$TrueBlock, SSA:$FalseBlock, CondClass:$Cond{CondClass::NEQ}, OpSize:$CompareSize{OpSize::iInvalid}, i1:$FromNZCV{false}": {
         "Inline": ["", "AddSub"],
         "HasSideEffects": true,
         "RAOverride": "2"
@@ -326,25 +313,13 @@
       "CallbackReturn": {
         "HasSideEffects": true
       },
-      "GPR = Syscall GPR:$SyscallID, GPR:$Arg0, GPR:$Arg1, GPR:$Arg2, GPR:$Arg3, GPR:$Arg4, GPR:$Arg5, SyscallFlags:$Flags": {
+      "GPR = Syscall GPR:$SyscallID, GPR:$Arg0, GPR:$Arg1, GPR:$Arg2, GPR:$Arg3, GPR:$Arg4, GPR:$Arg5": {
         "HasSideEffects": true,
         "Desc": ["Dispatches a guest syscall through to the SyscallHandler class"
                 ],
         "DestSize": "OpSize::i64Bit"
       },
 
-      "GPR = InlineSyscall GPR:$Arg0, GPR:$Arg1, GPR:$Arg2, GPR:$Arg3, GPR:$Arg4, GPR:$Arg5, i32:$HostSyscallNumber, SyscallFlags:$Flags": {
-        "HasSideEffects": true,
-        "Desc": ["Dispatches a guest syscall directly to the host syscall interface,",
-                 "bypassing the SyscallHandler class used by Syscall.",
-                 "This has significantly less overhead than Syscall, which needs to save JIT state first.",
-                 "Can only be used for syscalls that match across architecture,",
-                 "such as gettid (matches on x86/x86-64/Arm64)."
-                ],
-
-        "DestSize": "OpSize::i64Bit"
-      },
-
       "Thunk GPR:$ArgPtr, SHA256Sum:$ThunkNameHash": {
         "HasSideEffects": true
       },
@@ -364,20 +339,6 @@
       "GPR = Copy GPR:$Source": {
         "Desc": ["GPR copy, generated by RA to split live ranges"],
         "DestSize": "OpSize::i64Bit"
-      },
-
-      "GPR = Swap1 GPR:$A, GPR:$B": {
-        "Desc": ["GPR swap part 1, generated by RA. Returns value of first source.",
-                 "Destination must be second GPR."],
-        "DestSize": "OpSize::i64Bit"
-      },
-
-      "GPR = Swap2": {
-        "Desc": ["GPR swap part 2, generated by RA. Returns source source.",
-                 "Must immediately succeed Swap1 with no intervening instructions",
-                 "Kludge to workaround single destination restriction on IR",
-                 "Hopefully temporary"],
-        "DestSize": "OpSize::i64Bit"
       }
     },
     "StaticRA": {
@@ -423,8 +384,8 @@
                 ],
         "DestSize": "ByteSize",
         "EmitValidation": [
-          "($Class == GPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == GPRClass",
+          "($Class == RegClass::GPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == RegClass::FPR",
+          "($Class == RegClass::FPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == RegClass::GPR",
           "!($Offset >= offsetof(Core::CPUState, gregs[0]) && $Offset < offsetof(Core::CPUState, gregs[16])) && \"Can't LoadContext to GPR\"",
           "!($Offset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't LoadContext to XMM\""
         ]
@@ -437,8 +398,8 @@
         "HasSideEffects": true,
         "DestSize": "ByteSize",
         "EmitValidation": [
-          "($Class == GPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == GPRClass",
+          "($Class == RegClass::GPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == RegClass::FPR",
+          "($Class == RegClass::FPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == RegClass::GPR",
           "!($Offset >= offsetof(Core::CPUState, gregs[0]) && $Offset < offsetof(Core::CPUState, gregs[16])) && \"Can't LoadContext to GPR\"",
           "!($Offset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't LoadContext to XMM\""
         ]
@@ -454,8 +415,8 @@
         "HasSideEffects": true,
         "DestSize": "ByteSize",
         "EmitValidation": [
-          "($Class == GPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == GPRClass",
+          "($Class == RegClass::GPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == RegClass::FPR",
+          "($Class == RegClass::FPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == RegClass::GPR",
           "!($Offset >= offsetof(Core::CPUState, gregs[0]) && $Offset < offsetof(Core::CPUState, gregs[16])) && \"Can't StoreContext to GPR\"",
           "!($Offset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't StoreContext to XMM\""
         ]
@@ -472,8 +433,8 @@
         "EmitValidation": [
           "WalkFindRegClass($Value1) == $Class",
           "WalkFindRegClass($Value2) == $Class",
-          "($Class == GPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == GPRClass",
+          "($Class == RegClass::GPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == RegClass::FPR",
+          "($Class == RegClass::FPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == RegClass::GPR",
           "!($Offset >= offsetof(Core::CPUState, gregs[0]) && $Offset < offsetof(Core::CPUState, gregs[16])) && \"Can't StoreContext to GPR\"",
           "!($Offset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't StoreContext to XMM\""
         ]
@@ -485,8 +446,8 @@
                 ],
         "DestSize": "ByteSize",
         "EmitValidation": [
-          "($Class == GPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == GPRClass",
+          "($Class == RegClass::GPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == RegClass::FPR",
+          "($Class == RegClass::FPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == RegClass::GPR",
           "!($BaseOffset >= offsetof(Core::CPUState, gregs[0]) && $BaseOffset < offsetof(Core::CPUState, gregs[16])) && \"Can't LoadContextIndexed to GPR\"",
           "!($BaseOffset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $BaseOffset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't LoadContextIndexed to XMM\""
         ]
@@ -499,8 +460,8 @@
         "DestSize": "ByteSize",
         "EmitValidation": [
           "WalkFindRegClass($Value) == $Class",
-          "($Class == GPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == GPRClass",
+          "($Class == RegClass::GPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit)) || $Class == RegClass::FPR",
+          "($Class == RegClass::FPR && (#ByteSize == IR::OpSize::i8Bit || #ByteSize == IR::OpSize::i16Bit || #ByteSize == IR::OpSize::i32Bit || #ByteSize == IR::OpSize::i64Bit || #ByteSize == IR::OpSize::i128Bit || #ByteSize == IR::OpSize::i256Bit)) || $Class == RegClass::GPR",
           "!($BaseOffset >= offsetof(Core::CPUState, gregs[0]) && $BaseOffset < offsetof(Core::CPUState, gregs[16])) && \"Can't StoreContextIndexed to GPR\"",
           "!($BaseOffset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $BaseOffset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't StoreContextIndexed to XMM\""
         ]
@@ -630,7 +591,7 @@
         "DestSize": "RegisterSize",
         "ElementSize": "ElementSize"
       },
-      "FPR = VLoadVectorGatherMasked OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Incoming, FPR:$Mask, GPR:$AddrBase, FPR:$VectorIndexLow, FPR:$VectorIndexHigh, OpSize:$VectorIndexElementSize, u8:$OffsetScale, u8:$DataElementOffsetStart, u8:$IndexElementOffsetStart": {
+      "FPR = VLoadVectorGatherMasked OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Incoming, FPR:$Mask, GPR:$AddrBase, FPR:$VectorIndexLow, FPR:$VectorIndexHigh, OpSize:$VectorIndexElementSize, u8:$OffsetScale, u8:$DataElementOffsetStart, u8:$IndexElementOffsetStart, OpSize:$AddrSize": {
         "Desc": [
           "Does a masked load similar to VPGATHERD* where the upper bit of each element",
           "determines whether or not that element will be loaded from memory.",
@@ -644,7 +605,7 @@
           "$VectorIndexElementSize == OpSize::i32Bit || $VectorIndexElementSize == OpSize::i64Bit"
         ]
       },
-      "FPR = VLoadVectorGatherMaskedQPS OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Incoming, FPR:$MaskReg, GPR:$AddrBase, FPR:$VectorIndexLow, FPR:$VectorIndexHigh, u8:$OffsetScale": {
+      "FPR = VLoadVectorGatherMaskedQPS OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Incoming, FPR:$MaskReg, GPR:$AddrBase, FPR:$VectorIndexLow, FPR:$VectorIndexHigh, u8:$OffsetScale, OpSize:$AddrSize": {
         "Desc": [
           "Does a masked load similar to VPGATHERQPS where the upper bit of each element",
           "determines whether or not that element will be loaded from memory.",
@@ -759,9 +720,10 @@
       },
       "Fence FenceType:$Fence": {
         "Desc": ["Does a memory fence operation of the desired type",
-                 "Fence_Load: Ensures load memory operations are serialized",
-                 "Fence_Store: Ensures store memory operations are serialized",
-                 "Fence_LoadStore: Ensures loads and store memory operations are serialized",
+                 "FenceType::Load: Ensures load memory operations are serialized",
+                 "FenceType::Store: Ensures store memory operations are serialized",
+                 "FenceType::LoadStore: Ensures loads and store memory operations are serialized",
+                 "FenceType::Inst: Instruction barrier. Ensures all instructions after this point will be explicitly fetched",
                  "Ensures the memory operations are globally visible"
                 ],
         "HasSideEffects": true
@@ -842,16 +804,6 @@
           "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
-      "AtomicXor OpSize:#Size, GPR:$Value, GPR:$Addr": {
-        "HasSideEffects": true,
-        "Desc": ["Atomic integer xor",
-                 "IR layout must match Fetch-variant, otherwise DCE IR optimization breaks!"
-                ],
-        "DestSize": "Size",
-        "EmitValidation": [
-          "Size == FEXCore::IR::OpSize::i8Bit || Size == FEXCore::IR::OpSize::i16Bit || Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
-        ]
-      },
       "GPR = AtomicSwap OpSize:#Size, GPR:$Value, GPR:$Addr": {
         "HasSideEffects": true,
         "Desc": ["Atomic integer swap"
@@ -1006,7 +958,7 @@
         "DestSize": "OpSize::i64Bit"
       },
 
-      "GPR = Neg OpSize:#Size, GPR:$Src, CondClass:$Cond{{COND_AL}}": {
+      "GPR = Neg OpSize:#Size, GPR:$Src, CondClass:$Cond{CondClass::AL}": {
         "Desc": ["Integer negation, with optional predication",
                  "Dest = Cond ? -Src : Src",
                  "Will truncate to 64 or 32bits"
@@ -1084,6 +1036,13 @@
           "Size == FEXCore::IR::OpSize::i16Bit || Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
+      "GPR = Rbit OpSize:#Size, GPR:$Src": {
+        "Desc": ["Reverses the bit order of the register"],
+        "DestSize": "Size",
+        "EmitValidation": [
+          "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
+        ]
+      },
       "GPR = Add OpSize:#Size, GPR:$Src1, GPR:$Src2": {
         "Desc": [ "Integer Add",
                   "Will truncate to 64 or 32bits"
@@ -1551,6 +1510,15 @@
           "ResultSize == FEXCore::IR::OpSize::i32Bit || ResultSize == FEXCore::IR::OpSize::i64Bit"
         ]
       },
+      "GPR = MaskGenerateFromBitWidth GPR:$BitWidth": {
+        "Desc": ["Generates a bit mask from with a value from [0, 63]",
+                 "0 is special cased to full-mask",
+                 "Special operation for SSE4a bitmask generation."
+        ],
+        "DestSize": "FEXCore::IR::OpSize::i64Bit",
+        "ImplicitFlagClobber": true
+      },
+
       "GPR = Extr OpSize:#Size, GPR:$Upper, GPR:$Lower, u8:$LSB": {
         "Desc": ["Concats the two GPRs to create a value that is the size of the full two GPRs",
                  "It then extracts a bitfield width that size of a GPR from the LSB",
@@ -2159,22 +2127,34 @@
 
       "FPR = VAnd OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
         "DestSize": "RegisterSize",
-        "ElementSize": "ElementSize"
+        "ElementSize": "ElementSize",
+        "EmitValidation": [
+          "RegisterSize == FEXCore::IR::OpSize::i256Bit || RegisterSize == FEXCore::IR::OpSize::i128Bit || RegisterSize == FEXCore::IR::OpSize::i64Bit"
+        ]
       },
 
       "FPR = VAndn OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
         "DestSize": "RegisterSize",
-        "ElementSize": "ElementSize"
+        "ElementSize": "ElementSize",
+        "EmitValidation": [
+          "RegisterSize == FEXCore::IR::OpSize::i256Bit || RegisterSize == FEXCore::IR::OpSize::i128Bit || RegisterSize == FEXCore::IR::OpSize::i64Bit"
+        ]
       },
 
       "FPR = VOr OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
         "DestSize": "RegisterSize",
-        "ElementSize": "ElementSize"
+        "ElementSize": "ElementSize",
+        "EmitValidation": [
+          "RegisterSize == FEXCore::IR::OpSize::i256Bit || RegisterSize == FEXCore::IR::OpSize::i128Bit || RegisterSize == FEXCore::IR::OpSize::i64Bit"
+        ]
       },
 
       "FPR = VXor OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
         "DestSize": "RegisterSize",
-        "ElementSize": "ElementSize"
+        "ElementSize": "ElementSize",
+        "EmitValidation": [
+          "RegisterSize == FEXCore::IR::OpSize::i256Bit || RegisterSize == FEXCore::IR::OpSize::i128Bit || RegisterSize == FEXCore::IR::OpSize::i64Bit"
+        ]
       },
 
       "FPR = VUQAdd OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
@@ -2828,17 +2808,13 @@
         "X87": true,
         "HasSideEffects": true
       },
-      "PushStack FPR:$X80Src, SSA:$OriginalValue, OpSize:$LoadSize, i1:$Float": {
+      "PushStack FPR:$X80Src, FPR:$OriginalValue, OpSize:$LoadSize": {
         "Desc": [
           "Pushes the provided X80Src source on to the x87 stack.",
-          "Tracks OriginalValue as the original value of X80Src.",
+          "Tracks OriginalValue as the original value of X80Src. OriginalValue can be Invalid() in which case no tracking is done.",
           "Opsize is 128bit for F80 values, 64-bit for low precision.",
           "LoadSize the original load size, i.e. of size of OriginalValue.",
-          "Float: 80-bit, 64-bit, 32-bit",
-          "Int: 64-bit, 32-bit, 16-bit"
-        ],
-        "EmitValidation": [
-          "WalkFindRegClass($OriginalValue) == FPRClass || WalkFindRegClass($OriginalValue) == GPRClass"
+          "Float: 80-bit, 64-bit, 32-bit"
         ],
         "HasSideEffects": true,
         "X87": true
@@ -2850,13 +2826,12 @@
         "HasSideEffects": true,
         "X87": true
       },
-      "StoreStackMem OpSize:$SourceSize, OpSize:$StoreSize, GPR:$Addr, GPR:$Offset, OpSize:$Align, MemOffsetType:$OffsetType, u8:$OffsetScale, i1:$Float": {
+      "StoreStackMem OpSize:$SourceSize, OpSize:$StoreSize, GPR:$Addr, GPR:$Offset, OpSize:$Align, MemOffsetType:$OffsetType, u8:$OffsetScale": {
         "Desc": [
           "Takes the top value off the x87 stack and stores it to memory.",
           "SourceSize is 128bit for F80 values, 64-bit for low precision.",
           "StoreSize is the store size for conversion:",
-          "Float: 80-bit, 64-bit, or 32-bit",
-          "Int: 64-bit, 32-bit, 16-bit"
+          "Float: 80-bit, 64-bit, or 32-bit"
         ],
         "HasSideEffects": true,
         "X87": true

FEXCore/Source/Interface/IR/IRDumper.cpp

@@ -38,8 +38,8 @@ static void PrintArg(fextl::stringstream* out, const IRListView*, uint64_t Arg)
   *out << fextl::fmt::format("#{:#x}", Arg);
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, CondClassType Arg) {
-  if (Arg == COND_AL) {
+static void PrintArg(fextl::stringstream* out, const IRListView*, CondClass Arg) {
+  if (Arg == CondClass::AL) {
     *out << "ALWAYS";
     return;
   }
@@ -48,7 +48,7 @@ static void PrintArg(fextl::stringstream* out, const IRListView*, CondClassType
                                                                  "UGT", "ULE", "SGE",  "SLT", "SGT", "SLE", "TSTZ", "TSTNZ",
                                                                  "FLU", "FGE", "FLEU", "FGT", "FU",  "FNU"};
 
-  *out << CondNames[Arg];
+  *out << CondNames[FEXCore::ToUnderlying(Arg)];
 }
 
 static void PrintArg(fextl::stringstream* out, const IRListView*, MemOffsetType Arg) {
@@ -58,39 +58,39 @@ static void PrintArg(fextl::stringstream* out, const IRListView*, MemOffsetType
     "SXTW",
   };
 
-  *out << Names[Arg];
+  *out << Names[FEXCore::ToUnderlying(Arg)];
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, RegisterClassType Arg) {
-  if (Arg == GPRClass.Val) {
-    *out << "GPR";
-  } else if (Arg == GPRFixedClass.Val) {
-    *out << "GPRFixed";
-  } else if (Arg == FPRClass.Val) {
-    *out << "FPR";
-  } else if (Arg == FPRFixedClass.Val) {
-    *out << "FPRFixed";
-  } else {
-    *out << "Unknown Registerclass " << Arg;
-  }
+static void PrintArg(fextl::stringstream* out, const IRListView*, RegClass Arg) {
+  *out << [Arg] {
+    switch (Arg) {
+    case RegClass::Invalid: return "Invalid";
+    case RegClass::GPR: return "GPR";
+    case RegClass::GPRFixed: return "GPRFixed";
+    case RegClass::FPR: return "FPR";
+    case RegClass::FPRFixed: return "FPRFixed";
+    case RegClass::Complex: return "Complex";
+    }
+    return "<Unknown RegClass Type>";
+  }();
 }
 
 static void PrintArg(fextl::stringstream* out, const IRListView* IR, OrderedNodeWrapper Arg) {
   if (Arg.IsImmediate()) {
     auto PhyReg = PhysicalRegister(Arg);
 
-    switch (PhyReg.Class) {
-    case FEXCore::IR::GPRClass.Val: *out << "r"; break;
-    case FEXCore::IR::GPRFixedClass.Val: *out << "R"; break;
-    case FEXCore::IR::FPRClass.Val: *out << "v"; break;
-    case FEXCore::IR::FPRFixedClass.Val: *out << "V"; break;
-    case FEXCore::IR::ComplexClass.Val: *out << "c"; break;
-    case FEXCore::IR::InvalidClass.Val: *out << "invalid"; break;
+    switch (PhyReg.AsRegClass()) {
+    case RegClass::GPR: *out << "r"; break;
+    case RegClass::GPRFixed: *out << "R"; break;
+    case RegClass::FPR: *out << "v"; break;
+    case RegClass::FPRFixed: *out << "V"; break;
+    case RegClass::Complex: *out << "c"; break;
+    case RegClass::Invalid: *out << "invalid"; break;
     default: *out << "unknown"; break;
     }
 
-    if (PhyReg.Class != FEXCore::IR::InvalidClass.Val) {
-      *out << std::dec << (uint32_t)PhyReg.Reg;
+    if (PhyReg.AsRegClass() != RegClass::Invalid) {
+      *out << std::dec << uint32_t(PhyReg.Reg);
     }
 
     return;
@@ -124,41 +124,32 @@ static void PrintArg(fextl::stringstream* out, const IRListView* IR, OrderedNode
   }
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::FenceType Arg) {
-  if (Arg == IR::Fence_Load) {
-    *out << "Loads";
-  } else if (Arg == IR::Fence_Store) {
-    *out << "Stores";
-  } else if (Arg == IR::Fence_LoadStore) {
-    *out << "LoadStores";
-  } else {
-    *out << "<Unknown Fence Type>";
-  }
+static void PrintArg(fextl::stringstream* out, const IRListView*, FenceType Arg) {
+  *out << [Arg] {
+    switch (Arg) {
+    case FenceType::Load: return "Loads";
+    case FenceType::Store: return "Stores";
+    case FenceType::LoadStore: return "LoadStores";
+    case FenceType::Inst: return "Instruction";
+    }
+    return "<Unknown Fence Type>";
+  }();
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::RoundType Arg) {
-  switch (Arg) {
-  case FEXCore::IR::Round_Nearest: *out << "Nearest"; break;
-  case FEXCore::IR::Round_Negative_Infinity: *out << "-Inf"; break;
-  case FEXCore::IR::Round_Positive_Infinity: *out << "+Inf"; break;
-  case FEXCore::IR::Round_Towards_Zero: *out << "Towards Zero"; break;
-  case FEXCore::IR::Round_Host: *out << "Host"; break;
-  default: *out << "<Unknown Round Type>"; break;
-  }
+static void PrintArg(fextl::stringstream* out, const IRListView*, RoundMode Arg) {
+  *out << [Arg] {
+    switch (Arg) {
+    case RoundMode::Nearest: return "Nearest";
+    case RoundMode::NegInfinity: return "-Inf";
+    case RoundMode::PosInfinity: return "+Inf";
+    case RoundMode::TowardsZero: return "Towards Zero";
+    case RoundMode::Host: return "Host";
+    }
+    return "<Unknown Round Type>";
+  }();
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, 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;
-  }
-}
-
-static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::NamedVectorConstant Arg) {
+static void PrintArg(fextl::stringstream* out, const IRListView*, NamedVectorConstant Arg) {
   *out << [Arg] {
     // clang-format off
     switch (Arg) {
@@ -186,6 +177,22 @@ static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::N
         return "movmskps_shift";
       case NamedVectorConstant::NAMED_VECTOR_AESKEYGENASSIST_SWIZZLE:
         return "aeskeygenassist_swizzle";
+      case NamedVectorConstant::NAMED_VECTOR_BLENDPS_0110B:
+        return "blendps_0110b";
+      case NamedVectorConstant::NAMED_VECTOR_BLENDPS_0111B:
+        return "blendps_0111b";
+      case NamedVectorConstant::NAMED_VECTOR_BLENDPS_1001B:
+        return "blendps_1001b";
+      case NamedVectorConstant::NAMED_VECTOR_BLENDPS_1011B:
+        return "blendps_1011b";
+      case NamedVectorConstant::NAMED_VECTOR_BLENDPS_1101B:
+        return "blendps_1101b";
+      case NamedVectorConstant::NAMED_VECTOR_BLENDPS_1110B:
+        return "blendps_1110b";
+      case NamedVectorConstant::NAMED_VECTOR_MOVMASKB:
+        return "movmaskb";
+      case NamedVectorConstant::NAMED_VECTOR_MOVMASKB_UPPER:
+        return "movmaskb_upper";
       case NamedVectorConstant::NAMED_VECTOR_ZERO:
         return "vectorzero";
       case NamedVectorConstant::NAMED_VECTOR_X87_ONE:
@@ -216,9 +223,20 @@ static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::N
         return "cvtmax_i32";
       case NamedVectorConstant::NAMED_VECTOR_CVTMAX_I64:
         return "cvtmax_i64";
-      default:
-        return "<Unknown Named Vector Constant>";
+      case NamedVectorConstant::NAMED_VECTOR_F80_SIGN_MASK:
+        return "f80_sign_mask";
+      case NamedVectorConstant::NAMED_VECTOR_SHA1RNDS_K0:
+        return "sha1rnds_k0";
+      case NamedVectorConstant::NAMED_VECTOR_SHA1RNDS_K1:
+        return "sha1rnds_k1";
+      case NamedVectorConstant::NAMED_VECTOR_SHA1RNDS_K2:
+        return "sha1rnds_k2";
+      case NamedVectorConstant::NAMED_VECTOR_SHA1RNDS_K3:
+        return "sha1rnds_k3";
+      case NamedVectorConstant::NAMED_VECTOR_MAX:
+        return "<Programming Error: Printing MAX value>";
     }
+    return "<Unknown Named Vector Constant>";
     // clang-format on
   }();
 }
@@ -241,36 +259,43 @@ static void PrintArg(fextl::stringstream* out, const IRListView*, IndexNamedVect
       return "dppd_mask";
     case IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PBLENDW:
       return "pblendw";
-    default:
-      return "<Unknown Indexed Named Vector Constant>";
+    case INDEXED_NAMED_VECTOR_MAX:
+      return "<Programming Error: Printing MAX value>";
     }
+    return "<Unknown Indexed Named Vector Constant>";
     // clang-format on
   }();
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::OpSize Arg) {
-  switch (Arg) {
-  case OpSize::i8Bit: *out << "i8"; break;
-  case OpSize::i16Bit: *out << "i16"; break;
-  case OpSize::i32Bit: *out << "i32"; break;
-  case OpSize::i64Bit: *out << "i64"; break;
-  case OpSize::i128Bit: *out << "i128"; break;
-  case OpSize::i256Bit: *out << "i256"; break;
-  case OpSize::f80Bit: *out << "f80"; break;
-  default: *out << "<Unknown OpSize Type>"; break;
-  }
+static void PrintArg(fextl::stringstream* out, const IRListView*, OpSize Arg) {
+  *out << [Arg] {
+    switch (Arg) {
+    case OpSize::iUnsized: return "Unsized";
+    case OpSize::i8Bit: return "i8";
+    case OpSize::i16Bit: return "i16";
+    case OpSize::i32Bit: return "i32";
+    case OpSize::i64Bit: return "i64";
+    case OpSize::f80Bit: return "f80";
+    case OpSize::i128Bit: return "i128";
+    case OpSize::i256Bit: return "i256";
+    case OpSize::iInvalid: return "Invalid";
+    }
+    return "<Unknown OpSize Type>";
+  }();
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::FloatCompareOp Arg) {
-  switch (Arg) {
-  case FloatCompareOp::EQ: *out << "FEQ"; break;
-  case FloatCompareOp::LT: *out << "FLT"; break;
-  case FloatCompareOp::LE: *out << "FLE"; break;
-  case FloatCompareOp::UNO: *out << "UNO"; break;
-  case FloatCompareOp::NEQ: *out << "NEQ"; break;
-  case FloatCompareOp::ORD: *out << "ORD"; break;
-  default: *out << "<Unknown OpSize Type>"; break;
-  }
+static void PrintArg(fextl::stringstream* out, const IRListView*, FloatCompareOp Arg) {
+  *out << [Arg] {
+    switch (Arg) {
+    case FloatCompareOp::EQ: return "FEQ";
+    case FloatCompareOp::LT: return "FLT";
+    case FloatCompareOp::LE: return "FLE";
+    case FloatCompareOp::UNO: return "UNO";
+    case FloatCompareOp::NEQ: return "NEQ";
+    case FloatCompareOp::ORD: return "ORD";
+    }
+    return "<Unknown FloatCompareOp Type>";
+  }();
 }
 
 static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::BreakDefinition Arg) {
@@ -280,23 +305,28 @@ static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::B
   *out << static_cast<uint32_t>(Arg.si_code) << "}";
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::ShiftType Arg) {
-  switch (Arg) {
-  case ShiftType::LSL: *out << "LSL"; break;
-  case ShiftType::LSR: *out << "LSR"; break;
-  case ShiftType::ASR: *out << "ASR"; break;
-  case ShiftType::ROR: *out << "ROR"; break;
-  default: *out << "<Unknown Shift Type>"; break;
-  }
+static void PrintArg(fextl::stringstream* out, const IRListView*, ShiftType Arg) {
+  *out << [Arg] {
+    switch (Arg) {
+    case ShiftType::LSL: return "LSL";
+    case ShiftType::LSR: return "LSR";
+    case ShiftType::ASR: return "ASR";
+    case ShiftType::ROR: return "ROR";
+    }
+    return "<Unknown Shift Type>";
+  }();
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, FEXCore::IR::BranchHint Arg) {
-  switch (Arg) {
-  case BranchHint::None: *out << "None"; break;
-  case BranchHint::Call: *out << "Call"; break;
-  case BranchHint::Return: *out << "Return"; break;
-  default: *out << "<Unknown Branch Hint>"; break;
-  }
+static void PrintArg(fextl::stringstream* out, const IRListView*, BranchHint Arg) {
+  *out << [Arg] {
+    switch (Arg) {
+    case BranchHint::None: return "None";
+    case BranchHint::Call: return "Call";
+    case BranchHint::Return: return "Return";
+    case BranchHint::CheckTF: return "CheckTF";
+    }
+    return "<Unknown Branch Hint>";
+  }();
 }
 
 static void PrintArg(fextl::stringstream* out, const IRListView*, const std::array<uint8_t, 0x10>& Arg) {
@@ -323,8 +353,7 @@ void Dump(fextl::stringstream* out, const IRListView* IR) {
       auto BlockIROp = BlockHeader->C<FEXCore::IR::IROp_CodeBlock>();
 
       AddIndent();
-      *out << "(%" << IR->GetID(BlockNode) << ") "
-           << "CodeBlock ";
+      *out << "(%" << IR->GetID(BlockNode) << ") " << "CodeBlock ";
 
       *out << "%" << BlockIROp->Begin.ID() << ", ";
       *out << "%" << BlockIROp->Last.ID() << std::endl;
@@ -353,17 +382,17 @@ void Dump(fextl::stringstream* out, const IRListView* IR) {
 
           auto PhyReg = PhysicalRegister(CodeNode);
           if (!PhyReg.IsInvalid()) {
-            switch (PhyReg.Class) {
-            case FEXCore::IR::GPRClass.Val: *out << "(r"; break;
-            case FEXCore::IR::GPRFixedClass.Val: *out << "(R"; break;
-            case FEXCore::IR::FPRClass.Val: *out << "(v"; break;
-            case FEXCore::IR::FPRFixedClass.Val: *out << "(V"; break;
-            case FEXCore::IR::ComplexClass.Val: *out << "(complex"; break;
-            case FEXCore::IR::InvalidClass.Val: *out << "(invalid"; break;
+            switch (PhyReg.AsRegClass()) {
+            case RegClass::GPR: *out << "(r"; break;
+            case RegClass::GPRFixed: *out << "(R"; break;
+            case RegClass::FPR: *out << "(v"; break;
+            case RegClass::FPRFixed: *out << "(V"; break;
+            case RegClass::Complex: *out << "(complex"; break;
+            case RegClass::Invalid: *out << "(invalid"; break;
             default: *out << "(unknown"; break;
             }
-            if (PhyReg.Class != FEXCore::IR::InvalidClass.Val) {
-              *out << std::dec << (uint32_t)PhyReg.Reg << ")";
+            if (PhyReg.AsRegClass() != RegClass::Invalid) {
+              *out << std::dec << uint32_t(PhyReg.Reg) << ")";
             } else {
               *out << ")";
             }

FEXCore/Source/Interface/IR/IREmitter.cpp

@@ -33,14 +33,14 @@ bool IsBlockExit(FEXCore::IR::IROps Op) {
   }
 }
 
-FEXCore::IR::RegisterClassType IREmitter::WalkFindRegClass(Ref Node) {
+RegClass IREmitter::WalkFindRegClass(Ref Node) {
   auto Class = GetOpRegClass(Node);
   switch (Class) {
-  case GPRClass:
-  case FPRClass:
-  case GPRFixedClass:
-  case FPRFixedClass:
-  case InvalidClass: return Class;
+  case RegClass::GPR:
+  case RegClass::FPR:
+  case RegClass::GPRFixed:
+  case RegClass::FPRFixed:
+  case RegClass::Invalid: return Class;
   default: break;
   }
 
@@ -82,7 +82,7 @@ FEXCore::IR::RegisterClassType IREmitter::WalkFindRegClass(Ref Node) {
   }
   default: LOGMAN_MSG_A_FMT("Unhandled op type: {} {} in argument class validation", ToUnderlying(IROp->Op), GetOpName(Node)); break;
   }
-  return InvalidClass;
+  return RegClass::Invalid;
 }
 
 void IREmitter::ResetWorkingList() {

FEXCore/Source/Interface/IR/IREmitter.h

@@ -46,12 +46,12 @@ public:
    *
    * @{ */
 
-  FEXCore::IR::RegisterClassType WalkFindRegClass(Ref Node);
+  RegClass WalkFindRegClass(Ref Node);
 
   // These inlining helpers are used by IRDefines.inc so define first.
   Ref InlineMem(OpSize Size, Ref Offset, MemOffsetType OffsetType, uint8_t& OffsetScale, bool TSO = false) {
     uint64_t Imm {};
-    if (OffsetType != MEM_OFFSET_SXTX || !IsValueConstant(WrapNode(Offset), &Imm)) {
+    if (OffsetType != MemOffsetType::SXTX || !IsValueConstant(WrapNode(Offset), &Imm)) {
       return Offset;
     }
 
@@ -108,36 +108,86 @@ public:
   IRPair<IROp_Jump> _Jump() {
     return _Jump(InvalidNode);
   }
-  IRPair<IROp_CondJump> _CondJump(Ref ssa0, CondClassType cond = {COND_NEQ}) {
+  IRPair<IROp_CondJump> _CondJump(Ref ssa0, CondClass cond = CondClass::NEQ) {
     return _CondJump(ssa0, _Constant(0), InvalidNode, InvalidNode, cond, GetOpSize(ssa0));
   }
-  IRPair<IROp_CondJump> _CondJump(Ref ssa0, Ref ssa1, Ref ssa2, CondClassType cond = {COND_NEQ}) {
+  IRPair<IROp_CondJump> _CondJump(Ref ssa0, Ref ssa1, Ref ssa2, CondClass cond = CondClass::NEQ) {
     return _CondJump(ssa0, _Constant(0), ssa1, ssa2, cond, GetOpSize(ssa0));
   }
-  // TODO: Work to remove this implicit sized Select implementation.
-  IRPair<IROp_Select> _Select(uint8_t Cond, Ref ssa0, Ref ssa1, Ref ssa2, Ref ssa3, IR::OpSize CompareSize = OpSize::iUnsized) {
-    if (CompareSize == OpSize::iUnsized) {
-      CompareSize = std::max(OpSize::i32Bit, std::max(GetOpSize(ssa0), GetOpSize(ssa1)));
-    }
 
-    return _Select(std::max(OpSize::i32Bit, std::max(GetOpSize(ssa2), GetOpSize(ssa3))), CompareSize, CondClassType {Cond}, ssa0, ssa1, ssa2, ssa3);
+  IRPair<IROp_LoadContext> _LoadContextGPR(OpSize ByteSize, uint32_t Offset) {
+    return _LoadContext(ByteSize, RegClass::GPR, Offset);
   }
-  IRPair<IROp_LoadMem> _LoadMem(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref ssa0, IR::OpSize Align = OpSize::i8Bit) {
-    return _LoadMem(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
+  IRPair<IROp_LoadContext> _LoadContextFPR(OpSize ByteSize, uint32_t Offset) {
+    return _LoadContext(ByteSize, RegClass::FPR, Offset);
   }
-  IRPair<IROp_StoreMem> _StoreMem(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref Addr, Ref Value, IR::OpSize Align = OpSize::i8Bit) {
-    return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
+  IRPair<IROp_StoreContext> _StoreContextGPR(OpSize ByteSize, Ref Value, uint32_t Offset) {
+    return _StoreContext(ByteSize, RegClass::GPR, Value, Offset);
+  }
+  IRPair<IROp_StoreContext> _StoreContextFPR(OpSize ByteSize, Ref Value, uint32_t Offset) {
+    return _StoreContext(ByteSize, RegClass::FPR, Value, Offset);
   }
 
-  IRPair<IROp_Select> Select01(FEXCore::IR::OpSize CompareSize, CondClassType Cond, OrderedNode* Cmp1, OrderedNode* Cmp2) {
+  IRPair<IROp_LoadContextIndexed> _LoadContextGPRIndexed(Ref Index, OpSize ByteSize, uint32_t BaseOffset, uint32_t Stride) {
+    return _LoadContextIndexed(Index, ByteSize, BaseOffset, Stride, RegClass::GPR);
+  }
+  IRPair<IROp_LoadContextIndexed> _LoadContextFPRIndexed(Ref Index, OpSize ByteSize, uint32_t BaseOffset, uint32_t Stride) {
+    return _LoadContextIndexed(Index, ByteSize, BaseOffset, Stride, RegClass::FPR);
+  }
+  IRPair<IROp_StoreContextIndexed> _StoreContextGPRIndexed(Ref Value, Ref Index, OpSize ByteSize, uint32_t BaseOffset, uint32_t Stride) {
+    return _StoreContextIndexed(Value, Index, ByteSize, BaseOffset, Stride, RegClass::GPR);
+  }
+  IRPair<IROp_StoreContextIndexed> _StoreContextFPRIndexed(Ref Value, Ref Index, OpSize ByteSize, uint32_t BaseOffset, uint32_t Stride) {
+    return _StoreContextIndexed(Value, Index, ByteSize, BaseOffset, Stride, RegClass::FPR);
+  }
+
+  IRPair<IROp_LoadMem> _LoadMem(RegClass Class, OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMem(Class, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+  }
+  IRPair<IROp_LoadMem> _LoadMemGPR(OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMem(RegClass::GPR, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+  }
+  IRPair<IROp_LoadMem> _LoadMemGPR(OpSize Size, Ref Addr, Ref Offset, OpSize Align, MemOffsetType OffsetType, uint8_t OffsetScale) {
+    return _LoadMem(RegClass::GPR, Size, Addr, Offset, Align, OffsetType, OffsetScale);
+  }
+  IRPair<IROp_LoadMem> _LoadMemFPR(OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMem(RegClass::FPR, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+  }
+  IRPair<IROp_LoadMem> _LoadMemFPR(OpSize Size, Ref Addr, Ref Offset, OpSize Align, MemOffsetType OffsetType, uint8_t OffsetScale) {
+    return _LoadMem(RegClass::FPR, Size, Addr, Offset, Align, OffsetType, OffsetScale);
+  }
+  IRPair<IROp_StoreMem> _StoreMem(RegClass Class, OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
+  }
+  IRPair<IROp_StoreMem> _StoreMemGPR(OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMem(RegClass::GPR, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
+  }
+  IRPair<IROp_StoreMem> _StoreMemGPR(OpSize Size, Ref Value, Ref Addr, Ref Offset, OpSize Align, MemOffsetType OffsetType, uint8_t OffsetScale) {
+    return _StoreMem(RegClass::GPR, Size, Value, Addr, Offset, Align, OffsetType, OffsetScale);
+  }
+  IRPair<IROp_StoreMem> _StoreMemFPR(OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMem(RegClass::FPR, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
+  }
+  IRPair<IROp_StoreMem> _StoreMemFPR(OpSize Size, Ref Value, Ref Addr, Ref Offset, OpSize Align, MemOffsetType OffsetType, uint8_t OffsetScale) {
+    return _StoreMem(RegClass::FPR, Size, Value, Addr, Offset, Align, OffsetType, OffsetScale);
+  }
+
+  IRPair<IROp_StoreMemPair> _StoreMemPairGPR(OpSize Size, Ref Value1, Ref Value2, Ref Addr, uint32_t Offset) {
+    return _StoreMemPair(RegClass::GPR, Size, Value1, Value2, Addr, Offset);
+  }
+  IRPair<IROp_StoreMemPair> _StoreMemPairFPR(OpSize Size, Ref Value1, Ref Value2, Ref Addr, uint32_t Offset) {
+    return _StoreMemPair(RegClass::FPR, Size, Value1, Value2, Addr, Offset);
+  }
+
+  IRPair<IROp_Select> Select01(FEXCore::IR::OpSize CompareSize, CondClass Cond, OrderedNode* Cmp1, OrderedNode* Cmp2) {
     return _Select(OpSize::i64Bit, CompareSize, Cond, Cmp1, Cmp2, _InlineConstant(1), _InlineConstant(0));
   }
 
   IRPair<IROp_Select> To01(FEXCore::IR::OpSize CompareSize, OrderedNode* Cmp1) {
-    return Select01(CompareSize, CondClassType {COND_NEQ}, Cmp1, Constant(0));
+    return Select01(CompareSize, CondClass::NEQ, Cmp1, Constant(0));
   }
 
-  IRPair<IROp_NZCVSelect> _NZCVSelect01(CondClassType Cond) {
+  IRPair<IROp_NZCVSelect> _NZCVSelect01(CondClass Cond) {
     return _NZCVSelect(OpSize::i64Bit, Cond, _InlineConstant(1), _InlineConstant(0));
   }
 
@@ -250,7 +300,7 @@ public:
   }
 
   /**  @} */
-  FEXCore::IR::RegisterClassType WalkFindRegClass(OrderedNodeWrapper ssa) {
+  RegClass WalkFindRegClass(OrderedNodeWrapper ssa) {
     Ref RealNode = ssa.GetNode(DualListData.ListBegin());
     return WalkFindRegClass(RealNode);
   }

FEXCore/Source/Interface/IR/PassManager.h

@@ -39,7 +39,7 @@ public:
   }
 
 protected:
-  PassManager* Manager;
+  PassManager* Manager {};
 };
 
 class PassManager final {

FEXCore/Source/Interface/IR/Passes/IRValidation.cpp

@@ -93,11 +93,11 @@ void IRValidation::Run(IREmitter* IREmit) {
           // After RA, the destination needs to be assigned a register and class
           auto PhyReg = PhysicalRegister(CodeNode);
 
-          FEXCore::IR::RegisterClassType ExpectedClass = IR::GetRegClass(IROp->Op);
-          FEXCore::IR::RegisterClassType AssignedClass = FEXCore::IR::RegisterClassType {PhyReg.Class};
+          const auto ExpectedClass = IR::GetRegClass(IROp->Op);
+          const auto AssignedClass = PhyReg.AsRegClass();
 
           // If no register class was assigned
-          if (AssignedClass == IR::InvalidClass) {
+          if (AssignedClass == IR::RegClass::Invalid) {
             HadError |= true;
             Errors << "%" << ID << ": Had destination but with no register class assigned" << std::endl;
           }
@@ -109,10 +109,10 @@ void IRValidation::Run(IREmitter* IREmit) {
           }
 
           // Assigned class wasn't the expected class and it is a non-complex op
-          if (AssignedClass != ExpectedClass && ExpectedClass != IR::ComplexClass) {
+          if (AssignedClass != ExpectedClass && ExpectedClass != IR::RegClass::Complex) {
             HadWarning |= true;
-            Warnings << "%" << ID << ": Destination had register class " << AssignedClass.Val << " When register class "
-                     << ExpectedClass.Val << " Was expected" << std::endl;
+            Warnings << "%" << ID << ": Destination had register class " << uint32_t(AssignedClass) << " When register class "
+                     << uint32_t(ExpectedClass) << " Was expected" << std::endl;
           }
         }
       }

FEXCore/Source/Interface/IR/Passes/RedundantFlagCalculationElimination.cpp

@@ -2,22 +2,21 @@
 /*
 $info$
 tags: ir|opts
-desc: This is not used right now, possibly broken
 $end_info$
 */
 
-#include "FEXCore/Core/X86Enums.h"
-#include "FEXCore/Utils/CompilerDefs.h"
-#include "FEXCore/Utils/MathUtils.h"
-#include "FEXCore/fextl/deque.h"
 #include "Interface/IR/IR.h"
 #include "Interface/IR/IREmitter.h"
-
-#include <FEXCore/IR/IR.h>
-#include <FEXCore/Utils/Profiler.h>
-
 #include "Interface/IR/PassManager.h"
 
+#include <FEXCore/Core/X86Enums.h>
+#include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/CompilerDefs.h>
+#include <FEXCore/Utils/MathUtils.h>
+#include <FEXCore/Utils/Profiler.h>
+#include <FEXCore/fextl/deque.h>
+#include <FEXCore/fextl/vector.h>
+
 // Flag bit flags
 #define FLAG_V (1U << 0)
 #define FLAG_C (1U << 1)
@@ -62,36 +61,36 @@ struct FlagInfo {
     return {.Raw = R};
   }
 
-  bool Trivial() {
+  bool Trivial() const {
     return Raw == 0;
   }
 
-  unsigned Read() {
+  unsigned Read() const {
     return Bits(0, 8);
   }
 
-  unsigned Write() {
+  unsigned Write() const {
     return Bits(8, 8);
   }
 
-  bool CanEliminate() {
+  bool CanEliminate() const {
     return Bits(16, 1);
   }
 
-  bool Special() {
+  bool Special() const {
     return Bits(63, 1);
   }
 
-  IROps Replacement() {
+  IROps Replacement() const {
     return (IROps)Bits(32, 16);
   }
 
-  IROps ReplacementNoWrite() {
+  IROps ReplacementNoWrite() const {
     return (IROps)Bits(48, 16);
   }
 
 private:
-  unsigned Bits(unsigned Start, unsigned Count) {
+  unsigned Bits(unsigned Start, unsigned Count) const {
     return (Raw >> Start) & ((1u << Count) - 1);
   }
 };
@@ -154,45 +153,44 @@ public:
 
 private:
   FlagInfo Classify(IROp_Header* Node);
-  unsigned FlagForReg(unsigned Reg);
-  unsigned FlagsForCondClassType(CondClassType Cond);
+  unsigned FlagsForCondClassType(CondClass Cond);
   bool EliminateDeadCode(IREmitter* IREmit, Ref CodeNode, IROp_Header* IROp);
   void FoldBranch(IREmitter* IREmit, IRListView& CurrentIR, IROp_CondJump* Op, Ref CodeNode);
-  CondClassType X86ToArmFloatCond(CondClassType X86);
+  CondClass X86ToArmFloatCond(CondClass X86);
   bool ProcessBlock(IREmitter* IREmit, IRListView& CurrentIR, Ref Block, ControlFlowGraph& CFG);
   void OptimizeParity(IREmitter* IREmit, IRListView& CurrentIR, ControlFlowGraph& CFG);
 };
 
-unsigned DeadFlagCalculationEliminination::FlagsForCondClassType(CondClassType Cond) {
+unsigned DeadFlagCalculationEliminination::FlagsForCondClassType(CondClass Cond) {
   switch (Cond) {
-  case COND_AL: return 0;
+  case CondClass::AL: return 0;
 
-  case COND_MI:
-  case COND_PL: return FLAG_N;
+  case CondClass::MI:
+  case CondClass::PL: return FLAG_N;
 
-  case COND_EQ:
-  case COND_NEQ: return FLAG_Z;
+  case CondClass::EQ:
+  case CondClass::NEQ: return FLAG_Z;
 
-  case COND_UGE:
-  case COND_ULT: return FLAG_C;
+  case CondClass::UGE:
+  case CondClass::ULT: return FLAG_C;
 
-  case COND_VS:
-  case COND_VC:
-  case COND_FU:
-  case COND_FNU: return FLAG_V;
+  case CondClass::VS:
+  case CondClass::VC:
+  case CondClass::FU:
+  case CondClass::FNU: return FLAG_V;
 
-  case COND_UGT:
-  case COND_ULE: return FLAG_Z | FLAG_C;
+  case CondClass::UGT:
+  case CondClass::ULE: return FLAG_Z | FLAG_C;
 
-  case COND_SGE:
-  case COND_SLT:
-  case COND_FLU:
-  case COND_FGE: return FLAG_N | FLAG_V;
+  case CondClass::SGE:
+  case CondClass::SLT:
+  case CondClass::FLU:
+  case CondClass::FGE: return FLAG_N | FLAG_V;
 
-  case COND_SGT:
-  case COND_SLE:
-  case COND_FLEU:
-  case COND_FGT: return FLAG_N | FLAG_Z | FLAG_V;
+  case CondClass::SGT:
+  case CondClass::SLE:
+  case CondClass::FLEU:
+  case CondClass::FGT: return FLAG_N | FLAG_Z | FLAG_V;
 
   default: LOGMAN_THROW_A_FMT(false, "unknown cond class type"); return FLAG_NZCV;
   }
@@ -456,7 +454,7 @@ bool DeadFlagCalculationEliminination::EliminateDeadCode(IREmitter* IREmit, Ref
   return true;
 }
 
-CondClassType DeadFlagCalculationEliminination::X86ToArmFloatCond(CondClassType X86) {
+CondClass DeadFlagCalculationEliminination::X86ToArmFloatCond(CondClass X86) {
   // Table of x86 condition codes that map to arm64 condition codes, in the
   // sense that fcmp+axflag+branch(x86) is equivalent to fcmp+branch(arm).
   //
@@ -465,12 +463,12 @@ CondClassType DeadFlagCalculationEliminination::X86ToArmFloatCond(CondClassType
   //
   // SF/OF conditions are trivial and therefore shouldn't actually be generated
   switch (X86) {
-  case COND_UGE /* A  */: return {COND_FGE} /* GE */;
-  case COND_UGT /* AE */: return {COND_FGT} /* GT */;
-  case COND_ULT /* B  */: return {COND_SLT} /* LT */;
-  case COND_ULE /* BE */: return {COND_SLE} /* LE */;
-  case COND_SLE /* LE */: return {COND_SLE} /* LE */;
-  default: return {COND_AL};
+  case CondClass::UGE /* A  */: return CondClass::FGE /* GE */;
+  case CondClass::UGT /* AE */: return CondClass::FGT /* GT */;
+  case CondClass::ULT /* B  */: return CondClass::SLT /* LT */;
+  case CondClass::ULE /* BE */: return CondClass::SLE /* LE */;
+  case CondClass::SLE /* LE */: return CondClass::SLE /* LE */;
+  default: return CondClass::AL;
   }
 }
 
@@ -485,8 +483,8 @@ void DeadFlagCalculationEliminination::FoldBranch(IREmitter* IREmit, IRListView&
   auto Prev = CurrentIR.GetOp<IR::IROp_Header>(PrevWrap);
   if (Prev->Op == OP_AXFLAG) {
     // Pattern match a branch fed by AXFLAG.
-    CondClassType ArmCond = X86ToArmFloatCond(Op->Cond);
-    if (ArmCond == COND_AL) {
+    CondClass ArmCond = X86ToArmFloatCond(Op->Cond);
+    if (ArmCond == CondClass::AL) {
       return;
     }
 
@@ -495,7 +493,7 @@ void DeadFlagCalculationEliminination::FoldBranch(IREmitter* IREmit, IRListView&
     // Pattern match a branch fed by a compare. We could also handle bit tests
     // here, but tbz/tbnz has a limited offset range which we don't have a way to
     // deal with yet. Let's hope that's not a big deal.
-    if (!(Op->Cond == COND_NEQ || Op->Cond == COND_EQ) || (Prev->Size < OpSize::i32Bit)) {
+    if (!(Op->Cond == CondClass::NEQ || Op->Cond == CondClass::EQ) || (Prev->Size < OpSize::i32Bit)) {
       return;
     }
 
@@ -629,9 +627,9 @@ void DeadFlagCalculationEliminination::OptimizeParity(IREmitter* IREmit, IRListV
   }
 
   for (auto [Block, BlockHeader] : CurrentIR.GetBlocks()) {
-    auto ID = BlockHeader->C<IROp_CodeBlock>()->ID;
+    const auto ID = BlockHeader->C<IROp_CodeBlock>()->ID;
+    const auto& Predecessors = CFG.Get(ID)->Predecessors;
     bool Full = false;
-    auto Predecessors = CFG.Get(ID)->Predecessors;
 
     if (Predecessors.empty()) {
       // Conservatively assume there was full parity before the start block

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

@@ -12,6 +12,7 @@ $end_info$
 #include "Interface/IR/Passes.h"
 #include "Interface/Core/CPUID.h"
 #include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/fextl/vector.h>
@@ -22,7 +23,7 @@ using namespace FEXCore;
 
 namespace FEXCore::IR {
 namespace {
-  struct RegisterClass {
+  struct RegisterClassData {
     uint32_t Available;
     uint32_t Count;
 
@@ -32,9 +33,9 @@ namespace {
     Ref RegToSSA[32];
   };
 
-  IR::RegisterClassType GetRegClassFromNode(IR::IRListView* IR, IR::IROp_Header* IROp) {
-    IR::RegisterClassType Class = IR::GetRegClass(IROp->Op);
-    if (Class != IR::ComplexClass) {
+  IR::RegClass GetRegClassFromNode(IR::IRListView* IR, IR::IROp_Header* IROp) {
+    const auto Class = IR::GetRegClass(IROp->Op);
+    if (Class != IR::RegClass::Complex) {
       return Class;
     }
 
@@ -46,7 +47,7 @@ namespace {
     case IR::OP_LOADMEM:
     case IR::OP_LOADMEMTSO: return IROp->C<IR::IROp_LoadMem>()->Class;
     case IR::OP_FILLREGISTER: return IROp->C<IR::IROp_FillRegister>()->Class;
-    default: return IR::InvalidClass;
+    default: return IR::RegClass::Invalid;
     }
   };
 } // Anonymous namespace
@@ -56,15 +57,15 @@ public:
   explicit ConstrainedRAPass(const FEXCore::CPUIDEmu* CPUID)
     : CPUID {CPUID} {}
   void Run(IREmitter* IREmit) override;
-  void AddRegisters(IR::RegisterClassType Class, uint32_t RegisterCount) override;
+  void AddRegisters(IR::RegClass Class, uint32_t RegisterCount) override;
   bool TryPostRAMerge(Ref LastNode, Ref CodeNode, IROp_Header* IROp);
 
 private:
-  RegisterClass Classes[IR::NumClasses];
+  RegisterClassData Classes[IR::NumClasses];
 
-  IREmitter* IREmit;
-  IRListView* IR;
-  const FEXCore::CPUIDEmu* CPUID;
+  IREmitter* IREmit {};
+  IRListView* IR {};
+  const FEXCore::CPUIDEmu* CPUID {};
 
   // Map of nodes to their preferred register, to coalesce load/store reg.
   fextl::vector<PhysicalRegister> PreferredReg;
@@ -82,7 +83,7 @@ private:
   fextl::vector<bool> Seen;
 
   // SourcesNextUses is read backwards, this tracks the index
-  int64_t SourceIndex;
+  int64_t SourceIndex {};
 
   bool Rematerializable(IROp_Header* IROp) {
     return IROp->Op == OP_CONSTANT;
@@ -101,7 +102,7 @@ private:
     uint32_t SlotPlusOne = SpillSlots[IR->GetID(Node).Value];
     LOGMAN_THROW_A_FMT(SlotPlusOne >= 1, "Node must have been spilled");
 
-    RegisterClassType RegClass = GetRegClassFromNode(IR, IROp);
+    const auto RegClass = GetRegClassFromNode(IR, IROp);
     return IREmit->_FillRegister(IROp->Size, IROp->ElementSize, SlotPlusOne - 1, RegClass);
   };
 
@@ -109,7 +110,7 @@ private:
   // block, so we don't need to size the block up-front.
   fextl::vector<uint32_t> NextUses;
 
-  bool AnySpilled;
+  bool AnySpilled {};
 
   bool IsValidArg(OrderedNodeWrapper Arg) {
     if (Arg.IsInvalid()) {
@@ -120,7 +121,7 @@ private:
     return Op != OP_INLINECONSTANT && Op != OP_INLINEENTRYPOINTOFFSET;
   };
 
-  RegisterClass* GetClass(PhysicalRegister Reg) {
+  RegisterClassData* GetClass(PhysicalRegister Reg) {
     return &Classes[Reg.Class];
   };
 
@@ -133,13 +134,13 @@ private:
     LOGMAN_THROW_A_FMT(ID < SSAToReg.size(), "Only old nodes looked up");
 
     PhysicalRegister Reg = SSAToReg[ID];
-    RegisterClass* Class = GetClass(Reg);
+    RegisterClassData* Class = GetClass(Reg);
 
     return (Class->Available & GetRegBits(Reg)) == 0 && Class->RegToSSA[Reg.Reg] == Node;
   };
 
   void FreeReg(PhysicalRegister Reg) {
-    RegisterClass* Class = GetClass(Reg);
+    RegisterClassData* Class = GetClass(Reg);
     uint32_t RegBits = GetRegBits(Reg);
 
     LOGMAN_THROW_A_FMT(!(Class->Available & RegBits), "Register double-free");
@@ -187,22 +188,22 @@ private:
   };
 
   PhysicalRegister DecodeSRAReg(const IROp_Header* IROp, Ref Node) {
-    uint8_t FlagOffset = Classes[GPRFixedClass.Val].Count - 2;
+    uint8_t FlagOffset = Classes[FEXCore::ToUnderlying(RegClass::GPRFixed)].Count - 2;
 
     if (IROp->Op == OP_STOREREGISTER) {
       return PhysicalRegister(Node);
     } else if (IROp->Op == OP_LOADPF || IROp->Op == OP_STOREPF) {
-      return PhysicalRegister {GPRFixedClass, FlagOffset};
+      return PhysicalRegister {RegClass::GPRFixed, FlagOffset};
     } else if (IROp->Op == OP_LOADAF || IROp->Op == OP_STOREAF) {
-      return PhysicalRegister {GPRFixedClass, (uint8_t)(FlagOffset + 1)};
+      return PhysicalRegister {RegClass::GPRFixed, uint8_t(FlagOffset + 1)};
     } else {
       const IROp_LoadRegister* Op = IROp->C<IR::IROp_LoadRegister>();
 
-      LOGMAN_THROW_A_FMT(Op->Class == GPRClass || Op->Class == FPRClass, "SRA classes");
-      if (Op->Class == FPRClass) {
-        return PhysicalRegister {FPRFixedClass, (uint8_t)Op->Reg};
+      LOGMAN_THROW_A_FMT(Op->Class == RegClass::GPR || Op->Class == RegClass::FPR, "SRA classes");
+      if (Op->Class == RegClass::FPR) {
+        return PhysicalRegister {RegClass::FPRFixed, uint8_t(Op->Reg)};
       } else {
-        return PhysicalRegister {GPRFixedClass, (uint8_t)Op->Reg};
+        return PhysicalRegister {RegClass::GPRFixed, uint8_t(Op->Reg)};
       }
     }
   };
@@ -267,7 +268,7 @@ private:
     SourceIndex = SourcesNextUses.size();
   }
 
-  void SpillReg(RegisterClass* Class, IROp_CodeBlock* Block, IROp_Header* Exclude) {
+  void SpillReg(RegisterClassData* Class, IROp_CodeBlock* Block, IROp_Header* Exclude) {
     // We're about to use next-use information, so calculate it.
     if (!AnySpilled) {
       CalculateNextUses(Block, Exclude);
@@ -318,7 +319,7 @@ private:
     // If we already spilled the Candidate, we don't need to spill again.
     // Similarly, if we can rematerialize the instruction, we don't spill it.
     if (!Spilled && Header->Op != OP_CONSTANT) {
-      LOGMAN_THROW_A_FMT(Reg.Class == GetRegClassFromNode(IR, Header), "Consistent");
+      LOGMAN_THROW_A_FMT(Reg.AsRegClass() == GetRegClassFromNode(IR, Header), "Consistent");
 
       // SpillSlots allocation is deferred.
       if (SpillSlots.empty()) {
@@ -329,7 +330,7 @@ private:
       uint32_t Slot = IR->GetHeader()->SpillSlots++;
 
       // We must map here in case we're spilling something we shuffled.
-      auto SpillOp = IREmit->_SpillRegister(OrderedNodeWrapper::FromImmediate(Reg.Raw), Slot, RegisterClassType {Reg.Class});
+      auto SpillOp = IREmit->_SpillRegister(OrderedNodeWrapper::FromImmediate(Reg.Raw), Slot, Reg.AsRegClass());
       SpillOp.first->Header.Size = Header->Size;
       SpillOp.first->Header.ElementSize = Header->ElementSize;
       SpillSlots[Value] = Slot + 1;
@@ -341,7 +342,7 @@ private:
   };
 
   void RemapReg(Ref Node, PhysicalRegister Reg) {
-    RegisterClass* Class = GetClass(Reg);
+    RegisterClassData* Class = GetClass(Reg);
     Class->RegToSSA[Reg.Reg] = Node;
 
     uint32_t Index = IR->GetID(Node).Value;
@@ -352,7 +353,7 @@ private:
 
   // Record a given assignment of register Reg to Node.
   void SetReg(Ref Node, PhysicalRegister Reg) {
-    RegisterClass* Class = GetClass(Reg);
+    RegisterClassData* Class = GetClass(Reg);
     uint32_t RegBits = GetRegBits(Reg);
 
     LOGMAN_THROW_A_FMT((Class->Available & RegBits) == RegBits, "Precondition");
@@ -370,7 +371,7 @@ private:
     // Prioritize preferred registers.
     if (Node < PreferredReg.size()) {
       if (PhysicalRegister Reg = PreferredReg[Node]; !Reg.IsInvalid()) {
-        RegisterClass* Class = GetClass(Reg);
+        RegisterClassData* Class = GetClass(Reg);
         uint32_t RegBits = GetRegBits(Reg);
 
         if ((Class->Available & RegBits) == RegBits) {
@@ -383,10 +384,10 @@ private:
     // Try to handle tied registers. This can fail, the JIT will insert moves.
     if (int TiedIdx = IR::TiedSource(IROp->Op); TiedIdx >= 0) {
       auto Reg = PhysicalRegister(IROp->Args[TiedIdx]);
-      RegisterClass* Class = GetClass(Reg);
+      RegisterClassData* Class = GetClass(Reg);
       uint32_t RegBits = GetRegBits(Reg);
 
-      if (Reg.Class != GPRFixedClass && Reg.Class != FPRFixedClass && (Class->Available & RegBits) == RegBits) {
+      if (Reg.AsRegClass() != RegClass::GPRFixed && Reg.AsRegClass() != RegClass::FPRFixed && (Class->Available & RegBits) == RegBits) {
         SetReg(CodeNode, Reg);
         return;
       }
@@ -394,7 +395,7 @@ private:
 
     // Try to coalesce reserved pairs. Just a heuristic to remove some moves.
     if (IROp->Op == OP_ALLOCATEGPR && IROp->C<IROp_AllocateGPR>()->ForPair) {
-      uint32_t Available = Classes[GPRClass].Available;
+      uint32_t Available = Classes[FEXCore::ToUnderlying(RegClass::GPR)].Available;
 
       // Only choose base register R if R and R + 1 are both free
       Available &= (Available >> 1);
@@ -405,20 +406,20 @@ private:
 
       if (Available) {
         unsigned Reg = std::countr_zero(Available);
-        SetReg(CodeNode, PhysicalRegister(GPRClass, Reg));
+        SetReg(CodeNode, PhysicalRegister(RegClass::GPR, Reg));
         return;
       }
     } else if (IROp->Op == OP_ALLOCATEGPRAFTER) {
-      uint32_t Available = Classes[GPRClass].Available;
+      uint32_t Available = Classes[FEXCore::ToUnderlying(RegClass::GPR)].Available;
       auto After = PhysicalRegister(IROp->Args[0]);
       if ((After.Reg & 1) == 0 && Available & (1ull << (After.Reg + 1))) {
-        SetReg(CodeNode, PhysicalRegister(GPRClass, After.Reg + 1));
+        SetReg(CodeNode, PhysicalRegister(RegClass::GPR, After.Reg + 1));
         return;
       }
     }
 
-    RegisterClassType ClassType = GetRegClassFromNode(IR, IROp);
-    RegisterClass* Class = &Classes[ClassType];
+    RegClass ClassType = GetRegClassFromNode(IR, IROp);
+    RegisterClassData* Class = &Classes[FEXCore::ToUnderlying(ClassType)];
 
     // Spill to make room in the register file.
     if (!Class->Available) {
@@ -433,10 +434,10 @@ private:
   };
 };
 
-void ConstrainedRAPass::AddRegisters(IR::RegisterClassType Class, uint32_t RegisterCount) {
+void ConstrainedRAPass::AddRegisters(IR::RegClass Class, uint32_t RegisterCount) {
   LOGMAN_THROW_A_FMT(RegisterCount <= 31, "Up to 31 regs supported");
 
-  Classes[Class].Count = RegisterCount;
+  Classes[FEXCore::ToUnderlying(Class)].Count = RegisterCount;
 }
 
 inline bool KillMove(IROp_Header* LastOp, IROp_Header* IROp, Ref LastNode, Ref CodeNode) {
@@ -530,7 +531,7 @@ bool ConstrainedRAPass::TryPostRAMerge(Ref LastNode, Ref CodeNode, IROp_Header*
       const auto Result = CPUID->RunFunction(ConstantFunction, 0 /* leaf */);
 
       IREmit->SetWriteCursorBefore(CodeNode);
-      IREmit->_Fence({FEXCore::IR::Fence_Inst});
+      IREmit->_Fence(IR::FenceType::Inst);
       IREmit->_Constant(Result.eax).Node->Reg = PhysicalRegister(Op->OutEAX).Raw;
       IREmit->_Constant(Result.ebx).Node->Reg = PhysicalRegister(Op->OutEBX).Raw;
       IREmit->_Constant(Result.ecx).Node->Reg = PhysicalRegister(Op->OutECX).Raw;
@@ -663,7 +664,7 @@ void ConstrainedRAPass::Run(IREmitter* IREmit_) {
       // Static registers must be consistent at SRA load/store. Evict to ensure.
       if (auto Node = DecodeSRANode(IROp, CodeNode); Node != nullptr) {
         auto Reg = DecodeSRAReg(IROp, CodeNode);
-        RegisterClass* Class = &Classes[Reg.Class];
+        RegisterClassData* Class = &Classes[Reg.Class];
 
         if (!(Class->Available & (1u << Reg.Reg))) {
           Ref Old = Class->RegToSSA[Reg.Reg];
@@ -678,7 +679,7 @@ void ConstrainedRAPass::Run(IREmitter* IREmit_) {
 
             Ref Copy;
 
-            if (Reg.Class == FPRFixedClass) {
+            if (Reg.AsRegClass() == RegClass::FPRFixed) {
               IROp_Header* Header = IR->GetOp<IROp_Header>(Old);
               Copy = IREmit->_VMov(Header->Size, OrderedNodeWrapper::FromImmediate(Reg.Raw));
             } else {

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

@@ -12,14 +12,14 @@ $end_info$
 #include <stdint.h>
 
 namespace FEXCore::IR {
-struct RegisterClassType;
+enum class RegClass : uint32_t;
 
 class RegisterAllocationPass : public FEXCore::IR::Pass {
 public:
-  virtual void AddRegisters(FEXCore::IR::RegisterClassType Class, uint32_t RegisterCount) = 0;
+  virtual void AddRegisters(RegClass Class, uint32_t RegisterCount) = 0;
 
   // Number of GPRs usable for pairs at start of GPR set. Must be even.
-  uint32_t PairRegs;
+  uint32_t PairRegs {};
 };
 
 } // namespace FEXCore::IR

FEXCore/Source/Interface/IR/Passes/x87StackOptimizationPass.cpp

@@ -6,7 +6,6 @@
 #include "Interface/IR/PassManager.h"
 #include "FEXCore/IR/IR.h"
 #include "FEXCore/Utils/Profiler.h"
-#include "FEXCore/Utils/MathUtils.h"
 #include "FEXCore/Core/HostFeatures.h"
 #include "Interface/Core/Addressing.h"
 
@@ -66,7 +65,7 @@ public:
   int8_t TopOffset = 0;
 
   FixedSizeStack()
-    : buffer(FixedSizeStack::size, {StackSlot::UNUSED, T()}) {}
+    : buffer(FixedSizeStack::size, {StackSlot::UNUSED, T::Invalid}) {}
 
   void push(const T& Value) {
     rotate();
@@ -85,7 +84,7 @@ public:
   }
 
   void pop() {
-    buffer.front() = {StackSlot::INVALID, T()};
+    buffer.front() = {StackSlot::INVALID, T::Invalid};
     rotate(false);
   }
 
@@ -103,7 +102,7 @@ public:
 
   void clear() {
     for (auto& Elem : buffer) {
-      Elem = {StackSlot::UNUSED, T()};
+      Elem = {StackSlot::UNUSED, T::Invalid};
     }
     TopOffset = 0;
   }
@@ -171,28 +170,40 @@ private:
   // Helpers
   Ref RotateRight8(uint32_t V, Ref Amount);
 
-  void F80SplitStore_Helper(const IROp_StoreStackMem* Op, Ref StackNode) {
-    Ref AddrNode = IR->GetNode(Op->Addr);
-    Ref Offset = IR->GetNode(Op->Offset);
-    OpSize Align = Op->Align;
-    MemOffsetType OffsetType = Op->OffsetType;
-    uint8_t OffsetScale = Op->OffsetScale;
-
-    IREmit->_StoreMem(FPRClass, OpSize::i64Bit, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
+  void F80SplitStore_Helper(const IROp_StoreStackMem* Op, Ref StackNode, Ref AddrNode, Ref Offset, OpSize Align, MemOffsetType OffsetType,
+                            uint8_t OffsetScale) {
+    IREmit->_StoreMemFPR(OpSize::i64Bit, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
     auto Upper = IREmit->_VExtractToGPR(OpSize::i128Bit, OpSize::i64Bit, StackNode, 1);
 
     // Store the Upper part of the register (the remaining 2 bytes) into memory.
     AddressMode A {.Base = AddrNode,
                    .Index = Op->Offset.IsInvalid() ? nullptr : Offset,
-                   .IndexType = MEM_OFFSET_SXTX,
-                   .IndexScale = OffsetScale,
                    .Offset = 8,
+                   .IndexType = MemOffsetType::SXTX,
+                   .IndexScale = OffsetScale,
                    .AddrSize = OpSize::i64Bit};
     A = SelectAddressMode(IREmit, A, GPROpSize, Features.SupportsTSOImm9, false, false, OpSize::i16Bit);
-    IREmit->_StoreMem(GPRClass, OpSize::i16Bit, Upper, A.Base, A.Index, OpSize::i64Bit, MEM_OFFSET_SXTX, A.IndexScale);
+    IREmit->_StoreMemGPR(OpSize::i16Bit, Upper, A.Base, A.Index, OpSize::i64Bit, MemOffsetType::SXTX, A.IndexScale);
+  }
+
+  void Store80BitToMem(const IROp_StoreStackMem* Op, Ref StackNode, Ref AddrNode, Ref Offset, OpSize Align, MemOffsetType OffsetType,
+                       uint8_t OffsetScale) {
+    if (Features.SupportsSVE128 || Features.SupportsSVE256) {
+      AddressMode A {.Base = AddrNode,
+                     .Index = Op->Offset.IsInvalid() ? nullptr : Offset,
+                     .IndexType = MemOffsetType::SXTX,
+                     .IndexScale = OffsetScale,
+                     .AddrSize = OpSize::i64Bit};
+      AddrNode = LoadEffectiveAddress(IREmit, A, GPROpSize, false);
+      IREmit->_StoreMemX87SVEOptPredicate(OpSize::i128Bit, OpSize::i16Bit, StackNode, AddrNode);
+    } else {
+      F80SplitStore_Helper(Op, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
+    }
   }
 
   void StoreStackMem_Helper(const IROp_StoreStackMem* Op, Ref StackNode) {
+    LOGMAN_THROW_A_FMT(!ReducedPrecisionMode, "Full precision mode expected.");
+
     Ref AddrNode = IR->GetNode(Op->Addr);
     Ref Offset = IR->GetNode(Op->Offset);
     OpSize Align = Op->Align;
@@ -204,22 +215,12 @@ private:
     case OpSize::i32Bit:
     case OpSize::i64Bit: {
       StackNode = IREmit->_F80CVT(Op->StoreSize, StackNode);
-      IREmit->_StoreMem(FPRClass, Op->StoreSize, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
+      IREmit->_StoreMemFPR(Op->StoreSize, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
       break;
     }
 
     case OpSize::f80Bit: {
-      if (Features.SupportsSVE128 || Features.SupportsSVE256) {
-        AddressMode A {.Base = AddrNode,
-                       .Index = Op->Offset.IsInvalid() ? nullptr : Offset,
-                       .IndexType = MEM_OFFSET_SXTX,
-                       .IndexScale = OffsetScale,
-                       .AddrSize = OpSize::i64Bit};
-        AddrNode = LoadEffectiveAddress(IREmit, A, GPROpSize, false);
-        IREmit->_StoreMemX87SVEOptPredicate(OpSize::i128Bit, OpSize::i16Bit, StackNode, AddrNode);
-      } else { // 80bit requires split-store
-        F80SplitStore_Helper(Op, StackNode);
-      }
+      Store80BitToMem(Op, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
       break;
     }
     default: ERROR_AND_DIE_FMT("Unsupported x87 size");
@@ -229,6 +230,8 @@ private:
   // Performs a store to memory from a value the stack passed in as StackNode.
   // This is the version dealing with the reduced precision case.
   void StoreStackMem_Reduced_Helper(const IROp_StoreStackMem* Op, Ref StackNode) {
+    LOGMAN_THROW_A_FMT(ReducedPrecisionMode, "Reduced precision mode expected.");
+
     Ref AddrNode = IR->GetNode(Op->Addr);
     Ref Offset = IR->GetNode(Op->Offset);
     OpSize Align = Op->Align;
@@ -241,14 +244,13 @@ private:
       [[fallthrough]];
     }
     case OpSize::i64Bit: {
-      IREmit->_StoreMem(FPRClass, Op->StoreSize, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
+      IREmit->_StoreMemFPR(Op->StoreSize, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
       break;
     }
 
-    // 80bit requires split-store
     case OpSize::f80Bit: {
       StackNode = IREmit->_F80CVTTo(StackNode, OpSize::i64Bit);
-      F80SplitStore_Helper(Op, StackNode);
+      Store80BitToMem(Op, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
       break;
     }
     default: ERROR_AND_DIE_FMT("Unsupported x87 size");
@@ -290,23 +292,24 @@ private:
   void Reset();
 
   struct StackMemberInfo {
-    StackMemberInfo() {}
+    StackMemberInfo() = delete;
     StackMemberInfo(Ref Data)
       : StackDataNode(Data) {}
-    StackMemberInfo(Ref Data, Ref Source, OpSize Size, bool Float)
+    StackMemberInfo(Ref Data, Ref Source, OpSize Size)
       : StackDataNode(Data)
-      , Source({Size, Source})
-      , InterpretAsFloat(Float) {}
+      , Source({Size, Source}) {}
     Ref StackDataNode {}; // Reference to the data in the Stack.
                           // This is the source data node in the stack format, possibly converted to 64/80 bits.
     struct StackMemberData final {
       OpSize Size;
       Ref Node;
     };
+
+    static const StackMemberInfo Invalid;
+
     // Tuple is only valid if we have information about the Source of the Stack Data Node.
     // In it's valid then OpSize is the original source size and Ref is the original source node.
     std::optional<StackMemberData> Source {};
-    bool InterpretAsFloat {false}; // True if this is a floating point value, false if integer
   };
 
   // StackData, TopCache need to be always properly set to ensure
@@ -359,6 +362,8 @@ private:
   IRListView* IR = nullptr;
 };
 
+inline const X87StackOptimization::StackMemberInfo X87StackOptimization::StackMemberInfo::Invalid {nullptr};
+
 inline void X87StackOptimization::InvalidateCaches() {
   InvalidateCachedRegs();
   ConstantPool.fill(nullptr);
@@ -401,8 +406,7 @@ inline void X87StackOptimization::MigrateToSlowPathIf(bool ShouldMigrate) {
 
 inline Ref X87StackOptimization::GetTopWithCache_Slow() {
   if (!TopOffsetCache[0]) {
-    TopOffsetCache[0] =
-      IREmit->_LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
+    TopOffsetCache[0] = IREmit->_LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
   }
   return TopOffsetCache[0];
 }
@@ -447,7 +451,7 @@ inline void X87StackOptimization::SetTopWithCache_Slow(Ref Value) {
 
 inline Ref X87StackOptimization::GetFTW() {
   if (!FTWCached) {
-    FTWCached = IREmit->_LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+    FTWCached = IREmit->_LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
   }
   return FTWCached;
 }
@@ -470,7 +474,7 @@ inline Ref X87StackOptimization::LoadStackValueAtOffset_Slow(uint8_t Offset) {
   OrderedNode* TopOffsetAddress = GetOffsetTopAddressWithCache_Slow(Offset);
   auto Size = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
   if (!TopValueCache[Offset]) {
-    TopValueCache[Offset] = IREmit->_LoadMem(FPRClass, Size, TopOffsetAddress, IREmit->_InlineConstant(MMBaseOffset()), Size, MEM_OFFSET_SXTX, 1);
+    TopValueCache[Offset] = IREmit->_LoadMemFPR(Size, TopOffsetAddress, IREmit->_InlineConstant(MMBaseOffset()), Size, MemOffsetType::SXTX, 1);
   }
   return TopValueCache[Offset];
 }
@@ -595,7 +599,7 @@ inline void X87StackOptimization::UpdateTopForPush_Slow() {
 
 void X87StackOptimization::FlushCachedRegs() {
   if (FlushTopPending) {
-    IREmit->_StoreContext(OpSize::i8Bit, GPRClass, TopOffsetCache[0], offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
+    IREmit->_StoreContextGPR(OpSize::i8Bit, TopOffsetCache[0], offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
     FlushTopPending = false;
   }
 
@@ -603,7 +607,7 @@ void X87StackOptimization::FlushCachedRegs() {
   for (size_t i = 0; i < FlushValuesPending.size(); i++) {
     if (FlushValuesPending[i]) {
       OrderedNode* TopOffsetAddress = GetOffsetTopAddressWithCache_Slow(i);
-      IREmit->_StoreMem(FPRClass, Size, TopValueCache[i], TopOffsetAddress, IREmit->_InlineConstant(MMBaseOffset()), Size, MEM_OFFSET_SXTX, 1);
+      IREmit->_StoreMemFPR(Size, TopValueCache[i], TopOffsetAddress, IREmit->_InlineConstant(MMBaseOffset()), Size, MemOffsetType::SXTX, 1);
       // store
       FlushValuesPending[i] = false;
     }
@@ -654,7 +658,7 @@ void X87StackOptimization::FlushCachedRegs() {
         }
       }();
 
-      IREmit->_StoreContext(OpSize::i8Bit, GPRClass, NewFTW, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+      IREmit->_StoreContextGPR(OpSize::i8Bit, NewFTW, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
       FTWCached = NewFTW;
     }
 
@@ -729,6 +733,7 @@ void X87StackOptimization::Run(IREmitter* Emit) {
     // The optimization should run per-block
     Reset();
 
+    IREmit->SetCurrentCodeBlock(BlockNode);
     for (auto [CodeNode, IROp] : CurrentIR.GetCode(BlockNode)) {
       if (!LoweredX87(IROp->Op)) {
         continue;
@@ -928,8 +933,13 @@ void X87StackOptimization::Run(IREmitter* Emit) {
           StoreStackValueAtOffset_Slow(SourceNode);
         } else {
           auto* SourceNode = CurrentIR.GetNode(Op->X80Src);
-          auto* OriginalNode = CurrentIR.GetNode(Op->OriginalValue);
-          StackData.push(StackMemberInfo {SourceNode, OriginalNode, Op->LoadSize, Op->Float});
+          if (Op->OriginalValue.IsInvalid()) {
+            // No original value to track - just push the converted data
+            StackData.push(StackMemberInfo {SourceNode});
+          } else {
+            auto* OriginalNode = CurrentIR.GetNode(Op->OriginalValue);
+            StackData.push(StackMemberInfo {SourceNode, OriginalNode, Op->LoadSize});
+          }
         }
         break;
       }
@@ -994,9 +1004,16 @@ void X87StackOptimization::Run(IREmitter* Emit) {
         // str w2, [x1]
         // or similar. As long as the source size and dest size are one and the same.
         // This will avoid any conversions between source and stack element size and conversion back.
-        if (!SlowPath && Value->Source && Value->Source->Size == Op->StoreSize && Value->InterpretAsFloat) {
-          IREmit->_StoreMem(Value->InterpretAsFloat ? FPRClass : GPRClass, Op->StoreSize, Value->Source->Node, AddrNode, Offset, Align,
-                            OffsetType, OffsetScale);
+        OpSize StoreSize = Op->StoreSize;
+        LOGMAN_THROW_A_FMT(Op->StoreSize == OpSize::i32Bit || Op->StoreSize == OpSize::i64Bit || Op->StoreSize == OpSize::f80Bit,
+                           "Invalid store size in x87 store stack mem");
+        if (!SlowPath && Value->Source && Value->Source->Size == StoreSize) {
+          Ref SourceValue = Value->Source->Node;
+          if (Op->StoreSize == OpSize::f80Bit) {
+            Store80BitToMem(Op, SourceValue, AddrNode, Offset, Align, OffsetType, OffsetScale);
+          } else {
+            IREmit->_StoreMemFPR(StoreSize, SourceValue, AddrNode, Offset, Align, OffsetType, OffsetScale);
+          }
           break;
         }
 
@@ -1036,11 +1053,26 @@ void X87StackOptimization::Run(IREmitter* Emit) {
       case OP_F80STACKXCHANGE: {
         const auto* Op = IROp->C<IROp_F80StackXchange>();
         auto Offset = Op->SrcStack;
-        Ref ValueTop = LoadStackValue();
-        Ref ValueOffset = LoadStackValue(Offset);
 
-        StoreStackValue(ValueOffset);
-        StoreStackValue(ValueTop, Offset);
+        if (Offset == 0) {
+          // No-op
+          break;
+        }
+
+        const auto [ValidTop, StackMemberTop] = StackData.top(0);
+        const auto [ValidOffset, StackMemberOffset] = StackData.top(Offset);
+
+        if (ValidTop != StackSlot::VALID || ValidOffset != StackSlot::VALID) {
+          // Slow path: do actual memory operations
+          Ref ValueTop = LoadStackValue();
+          Ref ValueOffset = LoadStackValue(Offset);
+          StoreStackValue(ValueOffset);
+          StoreStackValue(ValueTop, Offset);
+        } else {
+          // Fast path: swap complete StackMemberInfo preserving Source metadata
+          StackData.setTop(StackMemberOffset, 0);
+          StackData.setTop(StackMemberTop, Offset);
+        }
         break;
       }
 
@@ -1160,7 +1192,7 @@ void X87StackOptimization::Run(IREmitter* Emit) {
 
         Ref Value {};
         if (ReducedPrecisionMode) {
-          Value = IREmit->_Vector_FToI(OpSize::i64Bit, OpSize::i64Bit, St0, Round_Host);
+          Value = IREmit->_Vector_FToI(OpSize::i64Bit, OpSize::i64Bit, St0, RoundMode::Host);
         } else {
           Value = IREmit->_F80Round(St0);
         }

FEXCore/Source/Interface/IR/RegisterAllocationData.h

@@ -19,9 +19,9 @@ union PhysicalRegister {
     return Raw == Other.Raw;
   }
 
-  PhysicalRegister(RegisterClassType Class, uint8_t Reg)
+  PhysicalRegister(RegClass Class, uint8_t Reg)
     : Reg(Reg)
-    , Class(Class.Val) {}
+    , Class(uint8_t(Class)) {}
 
   PhysicalRegister(OrderedNodeWrapper Arg)
     : Raw(Arg.GetImmediate()) {}
@@ -29,12 +29,16 @@ union PhysicalRegister {
   PhysicalRegister(Ref Node)
     : Raw(Node->Reg) {}
 
+  RegClass AsRegClass() const {
+    return RegClass {Class};
+  }
+
   static const PhysicalRegister Invalid() {
-    return PhysicalRegister(InvalidClass, 0);
+    return PhysicalRegister(RegClass::Invalid, 0);
   }
 
   bool IsInvalid() const {
-    static_assert(InvalidClass == 0);
+    static_assert(uint8_t(RegClass::Invalid) == 0);
     return Raw == 0;
   }
 };

FEXCore/Source/Utils/Allocator.cpp

@@ -4,6 +4,7 @@
 #include <FEXCore/Utils/CompilerDefs.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXCore/Utils/PrctlUtils.h>
 #include <FEXCore/Utils/TypeDefines.h>
 #include <FEXCore/fextl/fmt.h>
 #include <FEXCore/fextl/memory.h>
@@ -50,8 +51,17 @@ void* FEX_mmap(void* addr, size_t length, int prot, int flags, int fd, off_t off
     errno = -(uint64_t)Result;
     return (void*)-1;
   }
+
+  if (flags & MAP_ANONYMOUS) {
+    VirtualName("FEXMem", Result, length);
+  }
   return Result;
 }
+
+void VirtualName(const char* Name, void* Ptr, size_t Size) {
+  prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, Ptr, Size, Name);
+}
+
 int FEX_munmap(void* addr, size_t length) {
   int Result = Alloc64->Munmap(addr, length);
 
@@ -83,7 +93,7 @@ void* DisableSBRKAllocations() {
   // calls won't allocate any memory through that.
   void* AlignedBRK = reinterpret_cast<void*>(FEXCore::AlignUp(reinterpret_cast<uintptr_t>(StartingSBRK), FEXCore::Utils::FEX_PAGE_SIZE));
   void* AfterBRK =
-    mmap(AlignedBRK, FEXCore::Utils::FEX_PAGE_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED_NOREPLACE | MAP_NORESERVE, -1, 0);
+    ::mmap(AlignedBRK, FEXCore::Utils::FEX_PAGE_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED_NOREPLACE | MAP_NORESERVE, -1, 0);
   if (AfterBRK == INVALID_PTR) {
     // Couldn't allocate the page after the aligned brk? This should never happen.
     // FEXCore::LogMan isn't configured yet so we just need to print the message.
@@ -130,10 +140,7 @@ void ClearHooks() {
   FEXCore::Allocator::mmap = ::mmap;
   FEXCore::Allocator::munmap = ::munmap;
 
-  // XXX: This is currently a leak.
-  // We can't work around this yet until static initializers that allocate memory are completely removed from our codebase
-  // Luckily we only remove this on process shutdown, so the kernel will do the cleanup for us
-  Alloc64.release();
+  Alloc::OSAllocator::ReleaseAllocatorWorkaround(std::move(Alloc64));
 }
 #pragma GCC diagnostic pop
 
@@ -284,7 +291,7 @@ fextl::vector<MemoryRegion> StealMemoryRegion(uintptr_t Begin, uintptr_t End) {
       --StackRegionIt;
 
       auto Alloc =
-        mmap(StackRegionIt->Ptr, StackRegionIt->Size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED, -1, 0);
+        ::mmap(StackRegionIt->Ptr, StackRegionIt->Size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED, -1, 0);
 
       LogMan::Throw::AFmt(Alloc != MAP_FAILED, "mmap({},{:x}) failed", fmt::ptr(StackRegionIt->Ptr), StackRegionIt->Size);
       LogMan::Throw::AFmt(Alloc == StackRegionIt->Ptr, "mmap returned {} instead of {}", Alloc, fmt::ptr(StackRegionIt->Ptr));
@@ -295,7 +302,7 @@ fextl::vector<MemoryRegion> StealMemoryRegion(uintptr_t Begin, uintptr_t End) {
 
   // Block remaining memory gaps
   for (auto RegionIt = Regions.begin(); RegionIt != Regions.end(); ++RegionIt) {
-    auto Alloc = mmap(RegionIt->Ptr, RegionIt->Size, PROT_NONE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED_NOREPLACE, -1, 0);
+    auto Alloc = ::mmap(RegionIt->Ptr, RegionIt->Size, PROT_NONE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED_NOREPLACE, -1, 0);
 
     LogMan::Throw::AFmt(Alloc != MAP_FAILED, "mmap({},{:x}) failed", fmt::ptr(RegionIt->Ptr), RegionIt->Size);
     LogMan::Throw::AFmt(Alloc == RegionIt->Ptr, "mmap returned {} instead of {}", Alloc, fmt::ptr(RegionIt->Ptr));

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

@@ -98,7 +98,7 @@ private:
     // Align UsedPages so it pads to the next page.
     // Necessary to take advantage of madvise zero page pooling.
     using FlexBitElementType = uint64_t;
-    alignas(4096) FEXCore::FlexBitSet<FlexBitElementType> UsedPages;
+    alignas(FEXCore::Utils::FEX_PAGE_SIZE) FEXCore::FlexBitSet<FlexBitElementType> UsedPages;
 
     // This returns the size of the LiveVMARegion in addition to the flex set that tracks the used data
     // The LiveVMARegion lives at the start of the VMA region which means on initialization we need to set that
@@ -140,7 +140,7 @@ private:
     }
   };
 
-  static_assert(sizeof(LiveVMARegion) == 4096, "Needs to be the size of a page");
+  static_assert(sizeof(LiveVMARegion) == FEXCore::Utils::FEX_PAGE_SIZE, "Needs to be the size of a page");
 
   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");
@@ -168,6 +168,7 @@ private:
     LOGMAN_THROW_A_FMT(Res != -1, "Couldn't mprotect region: {} '{}' Likely occurs when running out of memory or Maximum VMAs", errno,
                        strerror(errno));
 
+    FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(ReservedRegion->Base), SizePlusManagedData);
     LiveVMARegion* LiveRange = new (reinterpret_cast<void*>(ReservedRegion->Base)) LiveVMARegion();
 
     // Copy over the reserved data
@@ -206,7 +207,7 @@ OSAllocator_64Bit::LiveVMARegion* OSAllocator_64Bit::FindLiveRegionForAddress(ui
     uintptr_t RegionBegin = (*it)->SlabInfo->Base;
     uintptr_t RegionEnd = RegionBegin + (*it)->SlabInfo->RegionSize;
 
-    if (Addr >= RegionBegin && Addr < RegionEnd) {
+    if (Addr >= RegionBegin && AddrEnd < RegionEnd) {
       LiveRegion = *it;
       // Leave our loop
       break;
@@ -404,14 +405,18 @@ again:
     // Mark the pages as used
     uintptr_t RegionBegin = LiveRegion->SlabInfo->Base;
     uintptr_t MappedBegin = (AllocatedOffset - RegionBegin) >> FEXCore::Utils::FEX_PAGE_SHIFT;
+    size_t PagesSet {};
 
     for (size_t i = 0; i < NumberOfPages; ++i) {
-      LiveRegion->UsedPages.Set(MappedBegin + i);
+      PagesSet += LiveRegion->UsedPages.TestAndSet(MappedBegin + i) == false;
     }
 
     // Change our last allocation region
     LiveRegion->LastPageAllocation = MappedBegin + NumberOfPages;
-    LiveRegion->FreeSpace -= length;
+    LiveRegion->FreeSpace -= PagesSet * FEXCore::Utils::FEX_PAGE_SIZE;
+    LOGMAN_THROW_A_FMT(LiveRegion->FreeSpace <= LiveRegion->SlabInfo->RegionSize,
+                       "Corrupt LiveRegion free space! 0x{:x} > 0x{:x}. After allocating 0x{:x} (0x{:x} overlapped)", LiveRegion->FreeSpace,
+                       LiveRegion->SlabInfo->RegionSize, length, PagesSet);
   }
 
   if (!AllocatedOffset) {
@@ -473,7 +478,7 @@ int OSAllocator_64Bit::Munmap(void* addr, size_t length) {
         ::mmap(addr, length, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
       }
 
-      (*it)->FreeSpace += FreedPages * 4096;
+      (*it)->FreeSpace += FreedPages * FEXCore::Utils::FEX_PAGE_SIZE;
 
       // Set the last allocated page to the minimum of last page allocation or this slab
       // This will let us more quickly fill holes
@@ -505,6 +510,8 @@ void OSAllocator_64Bit::AllocateMemoryRegions(fextl::vector<FEXCore::Allocator::
     // This enables the kernel to use transparent large pages in the allocator which can reduce memory pressure
     ::madvise(it.Ptr, ObjectAllocSize, MADV_HUGEPAGE);
 
+    FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(it.Ptr), ObjectAllocSize);
+
     ObjectAlloc = new (it.Ptr) Alloc::ForwardOnlyIntrusiveArenaAllocator(it.Ptr, ObjectAllocSize);
     ReservedRegions = ObjectAlloc->new_construct(ReservedRegions, ObjectAlloc);
     LiveRegions = ObjectAlloc->new_construct(LiveRegions, ObjectAlloc);
@@ -602,6 +609,8 @@ fextl::unique_ptr<T> make_alloc_unique(FEXCore::Allocator::MemoryRegion& Base, A
     ERROR_AND_DIE_FMT("Couldn't allocate memory region");
   }
 
+  FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(ptr), MinPage);
+
   // Remove the page from the base region.
   // Could be zero after this.
   Base.Size -= MinPage;

FEXCore/Source/Utils/Allocator/FlexBitSet.h

@@ -27,6 +27,11 @@ struct FlexBitSet final {
     Memory[Element / MinimumSizeBits] &= ~(1ULL << (Element % MinimumSizeBits));
     return Value;
   }
+  bool TestAndSet(size_t Element) {
+    bool Value = Get(Element);
+    Memory[Element / MinimumSizeBits] |= (1ULL << (Element % MinimumSizeBits));
+    return Value;
+  }
   void Set(size_t Element) {
     Memory[Element / MinimumSizeBits] |= (1ULL << (Element % MinimumSizeBits));
   }
@@ -70,12 +75,17 @@ struct FlexBitSet final {
   template<bool WantUnset>
   BitsetScanResults BackwardScanForRange(size_t BeginningElement, size_t ElementCount, size_t MinimumElement) {
     bool FoundHole {};
-    for (size_t CurrentPage = BeginningElement; CurrentPage >= (MinimumElement + ElementCount);) {
+
+    // Final element to iterate to.
+    const size_t FinalElement = MinimumElement + ElementCount - 1;
+
+    for (size_t CurrentPage = BeginningElement; CurrentPage >= FinalElement;) {
       size_t Remaining = ElementCount;
-      LOGMAN_THROW_A_FMT(Remaining <= CurrentPage, "Scanning less than available range");
+      LOGMAN_THROW_A_FMT(CurrentPage <= BeginningElement && CurrentPage >= FinalElement, "BackwardScanForRange: Scanning less than "
+                                                                                         "available range");
 
       while (Remaining) {
-        if (this->Get(CurrentPage - Remaining) == WantUnset) {
+        if (this->Get(CurrentPage - Remaining + 1) == WantUnset) {
           // Has an intersecting range
           break;
         }
@@ -92,7 +102,7 @@ struct FlexBitSet final {
         CurrentPage -= Remaining;
       } else {
         // We have a slab range
-        return BitsetScanResults {CurrentPage - ElementCount, FoundHole};
+        return BitsetScanResults {CurrentPage - ElementCount + 1, FoundHole};
       }
     }
 
@@ -108,11 +118,15 @@ struct FlexBitSet final {
   BitsetScanResults ForwardScanForRange(size_t BeginningElement, size_t ElementCount, size_t ElementsInSet) {
     bool FoundHole {};
 
-    for (size_t CurrentElement = BeginningElement; CurrentElement < (ElementsInSet - ElementCount);) {
+    // Final element to iterate to.
+    const size_t FinalElement = ElementsInSet - ElementCount + 1;
+
+    for (size_t CurrentElement = BeginningElement; CurrentElement <= FinalElement;) {
       // If we have enough free space, check if we have enough free pages that are contiguous
       size_t Remaining = ElementCount;
 
-      LOGMAN_THROW_A_FMT((CurrentElement + Remaining - 1) < ElementsInSet, "Scanning less than available range");
+      LOGMAN_THROW_A_FMT(CurrentElement >= BeginningElement && CurrentElement <= FinalElement, "ForwardScanForRange: Scanning less than "
+                                                                                               "available range");
 
       while (Remaining) {
         if (this->Get(CurrentElement + Remaining - 1) == WantUnset) {

FEXCore/Source/Utils/Allocator/HostAllocator.h

@@ -53,4 +53,12 @@ public:
 namespace Alloc::OSAllocator {
 fextl::unique_ptr<Alloc::HostAllocator> Create64BitAllocator();
 fextl::unique_ptr<Alloc::HostAllocator> Create64BitAllocatorWithRegions(fextl::vector<FEXCore::Allocator::MemoryRegion>& Regions);
+static inline void ReleaseAllocatorWorkaround(fextl::unique_ptr<Alloc::HostAllocator> Allocator) {
+  // XXX: This is currently a leak.
+  // We can't work around this yet until static initializers that allocate memory are completely removed from our codebase
+  // The allocator is also intrusively allocated, so the unique_ptr tries to double free the HostAllocator object.
+  // Luckily we only remove this on process shutdown, so the kernel will do the cleanup for us
+  Allocator.release();
+}
+
 } // namespace Alloc::OSAllocator

FEXCore/Source/Utils/ArchHelpers/Arm64.cpp

@@ -144,33 +144,33 @@ static __uint128_t LoadAcquire128(uint64_t Addr) {
 }
 
 static uint64_t LoadAcquire64(uint64_t Addr) {
-  std::atomic<uint64_t>* Atom = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
-  return Atom->load(std::memory_order_acquire);
+  auto Atom = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
+  return Atom.load(std::memory_order_acquire);
 }
 
 static bool StoreCAS64(uint64_t& Expected, uint64_t Val, uint64_t Addr) {
-  std::atomic<uint64_t>* Atom = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
-  return Atom->compare_exchange_strong(Expected, Val);
+  auto Atom = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
+  return Atom.compare_exchange_strong(Expected, Val);
 }
 
 static uint32_t LoadAcquire32(uint64_t Addr) {
-  std::atomic<uint32_t>* Atom = reinterpret_cast<std::atomic<uint32_t>*>(Addr);
-  return Atom->load(std::memory_order_acquire);
+  auto Atom = std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(Addr));
+  return Atom.load(std::memory_order_acquire);
 }
 
 static bool StoreCAS32(uint32_t& Expected, uint32_t Val, uint64_t Addr) {
-  std::atomic<uint32_t>* Atom = reinterpret_cast<std::atomic<uint32_t>*>(Addr);
-  return Atom->compare_exchange_strong(Expected, Val);
+  auto Atom = std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(Addr));
+  return Atom.compare_exchange_strong(Expected, Val);
 }
 
 static uint8_t LoadAcquire8(uint64_t Addr) {
-  std::atomic<uint8_t>* Atom = reinterpret_cast<std::atomic<uint8_t>*>(Addr);
-  return Atom->load(std::memory_order_acquire);
+  auto Atom = std::atomic_ref<uint8_t>(*reinterpret_cast<uint8_t*>(Addr));
+  return Atom.load(std::memory_order_acquire);
 }
 
 static bool StoreCAS8(uint8_t& Expected, uint8_t Val, uint64_t Addr) {
-  std::atomic<uint8_t>* Atom = reinterpret_cast<std::atomic<uint8_t>*>(Addr);
-  return Atom->compare_exchange_strong(Expected, Val);
+  auto Atom = std::atomic_ref<uint8_t>(*reinterpret_cast<uint8_t*>(Addr));
+  return Atom.compare_exchange_strong(Expected, Val);
 }
 
 static uint16_t DoLoad16(uint64_t Addr) {
@@ -211,8 +211,8 @@ static uint16_t DoLoad16(uint64_t Addr) {
         uint64_t Alignment = Addr & AlignmentMask;
         Addr &= ~AlignmentMask;
 
-        std::atomic<uint64_t>* Atomic = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
-        uint64_t TmpResult = Atomic->load();
+        auto Atomic = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
+        uint64_t TmpResult = Atomic.load();
 
         // Zexts the result
         uint16_t Result = TmpResult >> (Alignment * 8);
@@ -224,8 +224,8 @@ static uint16_t DoLoad16(uint64_t Addr) {
         uint64_t Alignment = Addr & AlignmentMask;
         Addr &= ~AlignmentMask;
 
-        std::atomic<uint32_t>* Atomic = reinterpret_cast<std::atomic<uint32_t>*>(Addr);
-        uint32_t TmpResult = Atomic->load();
+        auto Atomic = std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(Addr));
+        uint32_t TmpResult = Atomic.load();
 
         // Zexts the result
         uint16_t Result = TmpResult >> (Alignment * 8);
@@ -272,8 +272,8 @@ static uint32_t DoLoad32(uint64_t Addr) {
       uint64_t Alignment = Addr & AlignmentMask;
       Addr &= ~AlignmentMask;
 
-      std::atomic<uint64_t>* Atomic = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
-      uint64_t TmpResult = Atomic->load();
+      auto Atomic = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
+      uint64_t TmpResult = Atomic.load();
 
       return TmpResult >> (Alignment * 8);
     }
@@ -465,7 +465,7 @@ static bool RunCASPAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg1, uint3
       // Fits within a 16byte region
       uint64_t Alignment = Addr & 0b1111;
       Addr &= ~0b1111ULL;
-      std::atomic<__uint128_t>* Atomic128 = reinterpret_cast<std::atomic<__uint128_t>*>(Addr);
+      auto Atomic128 = std::atomic_ref<__uint128_t>(*reinterpret_cast<__uint128_t*>(Addr));
 
       __uint128_t Mask = ~0ULL;
       Mask <<= Alignment * 8;
@@ -480,7 +480,7 @@ static bool RunCASPAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg1, uint3
       Expected <<= Alignment * 8;
 
       while (1) {
-        TmpExpected = Atomic128->load();
+        TmpExpected = Atomic128.load();
 
         // Set up expected
         TmpExpected &= NegMask;
@@ -491,7 +491,7 @@ static bool RunCASPAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg1, uint3
         TmpDesired &= NegMask;
         TmpDesired |= Desired;
 
-        bool CASResult = Atomic128->compare_exchange_strong(TmpExpected, TmpDesired);
+        bool CASResult = Atomic128.compare_exchange_strong(TmpExpected, TmpDesired);
         if (CASResult) {
           // Successful, so we are done
           return true;
@@ -617,36 +617,6 @@ static uint64_t HandleCASPAL_ARMv8(uint32_t Instr, uintptr_t ProgramCounter, uin
   }
 }
 
-static bool HandleAtomicVectorStore(uint32_t Instr, uintptr_t ProgramCounter) {
-  uint32_t* PC = (uint32_t*)ProgramCounter;
-
-  uint32_t Size = (Instr >> 30) & 1;
-  uint32_t DataReg = Instr & 0x1F;
-
-  if (Size == 1) {
-    // 64-bit pair happens on paranoid vector stores
-    // [0] ldaxp(xzr, TMP3, MemSrc); // <- Can hit SIGBUS. Overwritten with DMB
-    // [1] stlxp(TMP3, TMP1, TMP2, MemSrc); // <- Can also hit SIGBUS
-    // [2] cbnz(TMP3, &B); // < Overwritten with DMB
-    if (DataReg == 31) {
-      uint32_t NextInstr = PC[1];
-      uint32_t AddrReg = (NextInstr >> 5) & 0x1F;
-      DataReg = NextInstr & 0x1F;
-      uint32_t DataReg2 = (NextInstr >> 10) & 0x1F;
-      uint32_t STP = (0b10 << 30) | (0b101001000000000 << 15) | (DataReg2 << 10) | (AddrReg << 5) | DataReg;
-
-      PC[0] = DMB;
-      PC[1] = STP;
-      PC[2] = DMB;
-      // Back up one instruction and have another go
-      ClearICache(&PC[0], 12);
-      return true;
-    }
-  }
-
-  return false;
-}
-
 template<typename T>
 using CASExpectedFn = T (*)(T Src, T Expected);
 template<typename T>
@@ -740,7 +710,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
       // Fits within a 16byte region
       uint64_t Alignment = Addr & 0b1111;
       Addr &= ~0b1111ULL;
-      std::atomic<__uint128_t>* Atomic128 = reinterpret_cast<std::atomic<__uint128_t>*>(Addr);
+      auto Atomic128 = std::atomic_ref<__uint128_t>(*reinterpret_cast<__uint128_t*>(Addr));
 
       __uint128_t Mask = 0xFFFF;
       Mask <<= Alignment * 8;
@@ -749,7 +719,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
       __uint128_t TmpDesired {};
 
       while (1) {
-        TmpExpected = Atomic128->load();
+        TmpExpected = Atomic128.load();
 
         __uint128_t Desired = DesiredFunction(TmpExpected >> (Alignment * 8), DesiredSrc);
         Desired <<= Alignment * 8;
@@ -766,7 +736,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
         TmpDesired &= NegMask;
         TmpDesired |= Desired;
 
-        bool CASResult = Atomic128->compare_exchange_strong(TmpExpected, TmpDesired);
+        bool CASResult = Atomic128.compare_exchange_strong(TmpExpected, TmpDesired);
         if (CASResult) {
           // Successful, so we are done
           return Expected >> (Alignment * 8);
@@ -810,9 +780,9 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
         uint64_t TmpExpected {};
         uint64_t TmpDesired {};
 
-        std::atomic<uint64_t>* Atomic = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
+        auto Atomic = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
         while (1) {
-          TmpExpected = Atomic->load();
+          TmpExpected = Atomic.load();
 
           uint64_t Desired = DesiredFunction(TmpExpected >> (Alignment * 8), DesiredSrc);
           Desired <<= Alignment * 8;
@@ -829,7 +799,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
           TmpDesired &= NegMask;
           TmpDesired |= Desired;
 
-          bool CASResult = Atomic->compare_exchange_strong(TmpExpected, TmpDesired);
+          bool CASResult = Atomic.compare_exchange_strong(TmpExpected, TmpDesired);
           if (CASResult) {
             // Successful, so we are done
             return Expected >> (Alignment * 8);
@@ -873,9 +843,9 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
         uint32_t TmpExpected {};
         uint32_t TmpDesired {};
 
-        std::atomic<uint32_t>* Atomic = reinterpret_cast<std::atomic<uint32_t>*>(Addr);
+        auto Atomic = std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(Addr));
         while (1) {
-          TmpExpected = Atomic->load();
+          TmpExpected = Atomic.load();
 
 
           uint32_t Desired = DesiredFunction(TmpExpected >> (Alignment * 8), DesiredSrc);
@@ -893,7 +863,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
           TmpDesired &= NegMask;
           TmpDesired |= Desired;
 
-          bool CASResult = Atomic->compare_exchange_strong(TmpExpected, TmpDesired);
+          bool CASResult = Atomic.compare_exchange_strong(TmpExpected, TmpDesired);
           if (CASResult) {
             // Successful, so we are done
             return Expected >> (Alignment * 8);
@@ -1040,7 +1010,7 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
       // Fits within a 16byte region
       uint64_t Alignment = Addr & 0b1111;
       Addr &= ~0b1111ULL;
-      std::atomic<__uint128_t>* Atomic128 = reinterpret_cast<std::atomic<__uint128_t>*>(Addr);
+      auto Atomic128 = std::atomic_ref<__uint128_t>(*reinterpret_cast<__uint128_t*>(Addr));
 
       __uint128_t Mask = ~0U;
       Mask <<= Alignment * 8;
@@ -1049,7 +1019,7 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
       __uint128_t TmpDesired {};
 
       while (1) {
-        __uint128_t TmpActual = Atomic128->load();
+        __uint128_t TmpActual = Atomic128.load();
 
         __uint128_t Desired = DesiredFunction(TmpActual >> (Alignment * 8), DesiredSrc);
         __uint128_t Expected = ExpectedFunction(TmpActual >> (Alignment * 8), ExpectedSrc);
@@ -1064,7 +1034,7 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
         TmpDesired &= NegMask;
         TmpDesired |= Desired << (Alignment * 8);
 
-        bool CASResult = Atomic128->compare_exchange_strong(TmpExpected, TmpDesired);
+        bool CASResult = Atomic128.compare_exchange_strong(TmpExpected, TmpDesired);
         if (CASResult) {
           // Stored successfully
           return Expected;
@@ -1108,9 +1078,9 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
       uint64_t TmpExpected {};
       uint64_t TmpDesired {};
 
-      std::atomic<uint64_t>* Atomic = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
+      auto Atomic = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
       while (1) {
-        uint64_t TmpActual = Atomic->load();
+        uint64_t TmpActual = Atomic.load();
 
         uint64_t Desired = DesiredFunction(TmpActual >> (Alignment * 8), DesiredSrc);
         uint64_t Expected = ExpectedFunction(TmpActual >> (Alignment * 8), ExpectedSrc);
@@ -1125,7 +1095,7 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
         TmpDesired &= NegMask;
         TmpDesired |= Desired << (Alignment * 8);
 
-        bool CASResult = Atomic->compare_exchange_strong(TmpExpected, TmpDesired);
+        bool CASResult = Atomic.compare_exchange_strong(TmpExpected, TmpDesired);
         if (CASResult) {
           // Stored successfully
           return Expected;
@@ -1270,7 +1240,7 @@ static uint64_t DoCAS64(uint64_t DesiredSrc, uint64_t ExpectedSrc, uint64_t Addr
     // Fits within a 16byte region
     uint64_t Alignment = Addr & AlignmentMask;
     Addr &= ~AlignmentMask;
-    std::atomic<__uint128_t>* Atomic128 = reinterpret_cast<std::atomic<__uint128_t>*>(Addr);
+    auto Atomic128 = std::atomic_ref<__uint128_t>(*reinterpret_cast<__uint128_t*>(Addr));
 
     __uint128_t Mask = ~0ULL;
     Mask <<= Alignment * 8;
@@ -1279,7 +1249,7 @@ static uint64_t DoCAS64(uint64_t DesiredSrc, uint64_t ExpectedSrc, uint64_t Addr
     __uint128_t TmpDesired {};
 
     while (1) {
-      __uint128_t TmpActual = Atomic128->load();
+      __uint128_t TmpActual = Atomic128.load();
 
       __uint128_t Desired = DesiredFunction(TmpActual >> (Alignment * 8), DesiredSrc);
       __uint128_t Expected = ExpectedFunction(TmpActual >> (Alignment * 8), ExpectedSrc);
@@ -1294,7 +1264,7 @@ static uint64_t DoCAS64(uint64_t DesiredSrc, uint64_t ExpectedSrc, uint64_t Addr
       TmpDesired &= NegMask;
       TmpDesired |= Desired << (Alignment * 8);
 
-      bool CASResult = Atomic128->compare_exchange_strong(TmpExpected, TmpDesired);
+      bool CASResult = Atomic128.compare_exchange_strong(TmpExpected, TmpDesired);
       if (CASResult) {
         // Stored successfully
         return Expected;
@@ -1326,9 +1296,7 @@ static uint64_t DoCAS64(uint64_t DesiredSrc, uint64_t ExpectedSrc, uint64_t Addr
   }
 }
 
-static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_t ExpectedReg, uint32_t AddressReg, uint32_t* StrictSplitLockMutex) {
-  uint64_t Addr = GPRs[AddressReg];
-
+static std::optional<uint64_t> DoCAS(uint32_t Size, uint64_t Desired, uint64_t Expected, uint64_t Addr, uint32_t* StrictSplitLockMutex) {
   // Cross-cacheline CAS doesn't work on ARM
   // It isn't even guaranteed to work on x86
   // Intel will do a "split lock" which locks the full bus
@@ -1341,7 +1309,7 @@ static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_
   // Only need to handle 16, 32, 64
   if (Size == 2) {
     auto Res = DoCAS16<false>(
-      GPRs[DesiredReg], GPRs[ExpectedReg], Addr,
+      Desired, Expected, Addr,
       [](uint16_t, uint16_t Expected) -> uint16_t {
         // Expected is just Expected
         return Expected;
@@ -1351,16 +1319,10 @@ static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_
         return Desired;
       },
       StrictSplitLockMutex);
-
-    // Regardless of pass or fail
-    // We set the result register if it isn't a zero register
-    if (ExpectedReg != 31) {
-      GPRs[ExpectedReg] = Res;
-    }
-    return true;
+    return Res;
   } else if (Size == 4) {
     auto Res = DoCAS32<false>(
-      GPRs[DesiredReg], GPRs[ExpectedReg], Addr,
+      Desired, Expected, Addr,
       [](uint32_t, uint32_t Expected) -> uint32_t {
         // Expected is just Expected
         return Expected;
@@ -1370,16 +1332,10 @@ static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_
         return Desired;
       },
       StrictSplitLockMutex);
-
-    // Regardless of pass or fail
-    // We set the result register if it isn't a zero register
-    if (ExpectedReg != 31) {
-      GPRs[ExpectedReg] = Res;
-    }
-    return true;
+    return Res;
   } else if (Size == 8) {
     auto Res = DoCAS64<false>(
-      GPRs[DesiredReg], GPRs[ExpectedReg], Addr,
+      Desired, Expected, Addr,
       [](uint64_t, uint64_t Expected) -> uint64_t {
         // Expected is just Expected
         return Expected;
@@ -1389,16 +1345,24 @@ static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_
         return Desired;
       },
       StrictSplitLockMutex);
-
-    // Regardless of pass or fail
-    // We set the result register if it isn't a zero register
-    if (ExpectedReg != 31) {
-      GPRs[ExpectedReg] = Res;
-    }
-    return true;
+    return Res;
   }
 
-  return false;
+  return std::nullopt;
+}
+
+static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_t ExpectedReg, uint32_t AddressReg, uint32_t* StrictSplitLockMutex) {
+  std::optional<uint64_t> Res = DoCAS(Size, GPRs[DesiredReg], GPRs[ExpectedReg], GPRs[AddressReg], StrictSplitLockMutex);
+  if (!Res.has_value()) {
+    return false;
+  }
+
+  // Regardless of pass or fail
+  // We set the result register if it isn't a zero register
+  if (ExpectedReg != 31) {
+    GPRs[ExpectedReg] = *Res;
+  }
+  return true;
 }
 
 static bool HandleCASAL(uint64_t* GPRs, uint32_t Instr, uint32_t* StrictSplitLockMutex) {
@@ -1560,38 +1524,43 @@ static bool HandleAtomicMemOp(uint32_t Instr, uint64_t* GPRs, uint32_t* StrictSp
   return false;
 }
 
-static bool HandleAtomicLoad(uint32_t Instr, uint64_t* GPRs, int64_t Offset) {
+static bool HandleAtomicLoad(uint32_t Instr, uint64_t* GPRs, int64_t Offset, Core::UnalignedExclusiveStore* Store = nullptr) {
   uint32_t Size = 1 << (Instr >> 30);
 
   uint32_t ResultReg = Instr & 0b11111;
   uint32_t AddressReg = (Instr >> 5) & 0b11111;
 
   uint64_t Addr = GPRs[AddressReg] + Offset;
+  uint64_t Res;
 
   if (Size == 2) {
-    auto Res = DoLoad16(Addr);
+    Res = DoLoad16(Addr);
     // We set the result register if it isn't a zero register
     if (ResultReg != 31) {
       GPRs[ResultReg] = Res;
     }
-    return true;
   } else if (Size == 4) {
-    auto Res = DoLoad32(Addr);
+    Res = DoLoad32(Addr);
     // We set the result register if it isn't a zero register
     if (ResultReg != 31) {
       GPRs[ResultReg] = Res;
     }
-    return true;
   } else if (Size == 8) {
-    auto Res = DoLoad64(Addr);
+    Res = DoLoad64(Addr);
     // We set the result register if it isn't a zero register
     if (ResultReg != 31) {
       GPRs[ResultReg] = Res;
     }
-    return true;
+  } else {
+    return false;
   }
 
-  return false;
+  if (Store) {
+    Store->Addr = Addr;
+    Store->Store = Res;
+    Store->Size = Size;
+  }
+  return true;
 }
 
 static bool HandleAtomicStore(uint32_t Instr, uint64_t* GPRs, int64_t Offset, uint32_t* StrictSplitLockMutex) {
@@ -1952,8 +1921,8 @@ static uint64_t HandleAtomicLoadstoreExclusive(uintptr_t ProgramCounter, uint64_
 }
 
 [[nodiscard]]
-std::optional<int32_t>
-HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandlerType HandleType, uintptr_t ProgramCounter, uint64_t* GPRs) {
+std::optional<int32_t> HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandlerType HandleType,
+                                             uintptr_t ProgramCounter, uint64_t* GPRs, bool IsJIT) {
 #ifdef _M_ARM_64
   constexpr bool is_arm64 = true;
 #else
@@ -1977,8 +1946,7 @@ HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandl
   auto CTX = static_cast<Context::ContextImpl*>(Thread->CTX);
   uint32_t* StrictSplitLockMutex {CTX->Config.StrictInProcessSplitLocks ? &CTX->StrictSplitLockMutex : nullptr};
 
-  // ParanoidTSO path doesn't modify any code.
-  if (HandleType == UnalignedHandlerType::Paranoid) [[unlikely]] {
+  if (!IsJIT) [[unlikely]] {
     if ((Instr & LDAXR_MASK) == LDAR_INST ||  // LDAR*
         (Instr & LDAXR_MASK) == LDAPR_INST) { // LDAPR*
       if (ArchHelpers::Arm64::HandleAtomicLoad(Instr, GPRs, 0)) {
@@ -2016,7 +1984,29 @@ HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandl
         LogMan::Msg::EFmt("Unhandled JIT SIGBUS LDLUR*: PC: 0x{:x} Instruction: 0x{:08x}\n", ProgramCounter, PC[0]);
         return std::nullopt;
       }
+    } else if ((Instr & ArchHelpers::Arm64::LDAXR_MASK) == ArchHelpers::Arm64::LDAXR_INST) { // LDAXR*
+      if (ArchHelpers::Arm64::HandleAtomicLoad(Instr, GPRs, 0, &Thread->ExclusiveStore)) {
+        return 4;
+      }
+    } else if ((Instr & ArchHelpers::Arm64::STLXR_MASK) == ArchHelpers::Arm64::STLXR_INST) { // STLXR*
+      uint32_t StatusReg = Instr << 11 >> 27;
+      // // Emulate exclusive store by validating the address and value against the last unaligned LDAXR*.
+      if (GPRs[AddrReg] != Thread->ExclusiveStore.Addr || Size > Thread->ExclusiveStore.Size) {
+        if (StatusReg != 31) {
+          GPRs[StatusReg] = 1;
+        }
+        return 4;
+      }
+      if (std::optional<uint64_t> Prev =
+            DoCAS(Size, DataReg == 31 ? 0 : GPRs[DataReg], Thread->ExclusiveStore.Store, GPRs[AddrReg], StrictSplitLockMutex)) {
+        if (StatusReg != 31) {
+          GPRs[StatusReg] = !!memcmp(&Thread->ExclusiveStore.Store, &*Prev, Size);
+        }
+        Thread->ExclusiveStore.Size = 0;
+        return 4;
+      }
     }
+    return 0;
   }
 
   const auto Frame = Thread->CurrentFrame;
@@ -2068,6 +2058,9 @@ HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandl
     if (BytesToSkip) {
       // Skip this instruction now
       return BytesToSkip;
+    } else {
+      LogMan::Msg::EFmt("Unhandled JIT SIGBUS CASPAL: PC: 0x{:x} Instruction: 0x{:08x}\n", ProgramCounter, PC[0]);
+      return std::nullopt;
     }
   }
 
@@ -2131,19 +2124,6 @@ HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandl
     ClearICache(&PC[-1], 8);
     // Back up one instruction and have another go
     return -4;
-  } else if ((Instr & ArchHelpers::Arm64::LDAXP_MASK) == ArchHelpers::Arm64::LDAXP_INST) { // LDAXP
-    /// This is handling the case of paranoid ARMv8.0-a atomic stores.
-    /// This backpatches the ldaxp+stlxp+cbnz if the previous `HandleCASPAL_ARMv8` didn't handle the case.
-    if (ArchHelpers::Arm64::HandleAtomicVectorStore(Instr, ProgramCounter)) {
-      return 0;
-    } else {
-      LogMan::Msg::EFmt("Unhandled JIT SIGBUS LDAXP: PC: 0x{:x} Instruction: 0x{:08x}\n", ProgramCounter, PC[0]);
-      return std::nullopt;
-    }
-  } else if ((Instr & ArchHelpers::Arm64::STLXP_MASK) == ArchHelpers::Arm64::STLXP_INST) { // STLXP
-    // Should not trigger - middle of an LDAXP/STAXP pair.
-    LogMan::Msg::EFmt("Unhandled JIT SIGBUS STLXP: PC: 0x{:x} Instruction: 0x{:08x}\n", ProgramCounter, PC[0]);
-    return std::nullopt;
   }
 
   // Check if another thread backpatched this instruction before this thread got here

FEXCore/Source/Utils/LongJump.cpp

@@ -1,7 +1,10 @@
 // SPDX-License-Identifier: MIT
 #include <FEXCore/Utils/LongJump.h>
+#include <FEXCore/Utils/LogManager.h>
 
-namespace FEXCore::LongJump {
+#include <cstring>
+
+namespace FEXCore::UncheckedLongJump {
 #if defined(_M_ARM_64)
 [[nodiscard]]
 FEX_DEFAULT_VISIBILITY FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
@@ -32,7 +35,7 @@ FEX_DEFAULT_VISIBILITY FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
 }
 
 [[noreturn]]
-FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value) {
+FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(const JumpBuf& Buffer, uint64_t Value) {
   __asm volatile(R"(
       // x0 contains the jumpbuffer
       ldp x19, x20, [x0, #( 0 * 8)];
@@ -58,6 +61,27 @@ FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value)
     )" ::
                    : "memory");
 }
+
+FEX_DEFAULT_VISIBILITY void ManuallyLoadJumpBuf(const JumpBuf& Buffer, uint64_t Value, uint64_t* GPRs, __uint128_t* FPRs, uint64_t* PC) {
+  // First 12 values are registers [x19,x30].
+  memcpy(&GPRs[19], &Buffer.Registers[0], sizeof(uint64_t) * 12);
+
+  // Next 8 values are [D8,D15]
+  // Retain upper 64-bits of the register, only modifying lower 64-bits.
+  for (size_t i = 0; i < 8; ++i) {
+    memcpy(&FPRs[8 + i], &Buffer.Registers[12 + i], sizeof(uint64_t));
+  }
+
+  // Last value is stack pointer
+  memcpy(&GPRs[31], &Buffer.Registers[20], sizeof(uint64_t));
+
+  // Load the expected value in to X0
+  GPRs[0] = Value;
+
+  // Load the PC with the current LR.
+  *PC = GPRs[30];
+}
+
 #else
 [[nodiscard]]
 FEX_DEFAULT_VISIBILITY FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
@@ -86,7 +110,7 @@ FEX_DEFAULT_VISIBILITY FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
 }
 
 [[noreturn]]
-FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value) {
+FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(const JumpBuf& Buffer, uint64_t Value) {
   __asm volatile(R"(
     .intel_syntax noprefix;
     // rdi contains the jumpbuffer
@@ -115,5 +139,9 @@ FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value)
                    : "memory");
 }
 
+FEX_DEFAULT_VISIBILITY void ManuallyLoadJumpBuf(JumpBuf& Buffer, uint64_t Value, uint64_t* GPRs, __uint128_t* FPRs, uint64_t* PC) {
+  LOGMAN_MSG_A_FMT("This is unimplemented on x86-64");
+}
+
 #endif
-} // namespace FEXCore::LongJump
+} // namespace FEXCore::UncheckedLongJump

FEXCore/Source/Utils/Profiler.cpp

@@ -9,6 +9,7 @@
 
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Profiler.h>
+#include <FEXCore/Config/Config.h>
 #include <FEXCore/fextl/fmt.h>
 #include <FEXCore/fextl/string.h>
 
@@ -70,6 +71,10 @@ static std::array<const char*, 2> TraceFSDirectories {
 };
 
 void Init() {
+  FEX_CONFIG_OPT(EnableGpuvisProfiling, ENABLEGPUVISPROFILING);
+  if (!EnableGpuvisProfiling()) {
+    return;
+  }
   for (auto Path : TraceFSDirectories) {
 #ifdef _WIN32
     constexpr auto flags = O_WRONLY;

FEXCore/Source/Utils/SpinWaitLock.h

@@ -1,9 +1,14 @@
 // SPDX-License-Identifier: MIT
+#pragma once
+
 #include <atomic>
 #include <chrono>
 #include <mutex>
 #include <type_traits>
 
+#include <FEXCore/fextl/functional.h>
+#include <FEXCore/Utils/EnumUtils.h>
+
 namespace FEXCore::Utils::SpinWaitLock {
 /**
  * @brief This provides routines to implement implement an "efficient spin-loop" using ARM's WFE and exclusive monitor interfaces.
@@ -123,35 +128,26 @@ static inline uint64_t WFELoadAtomic(uint64_t* Futex) {
   return Result;
 }
 
-template<typename T, typename TT = T>
-static inline void Wait(T* Futex, TT ExpectedValue) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
-  T Result = AtomicFutex->load();
+template<typename Pred, typename T>
+static inline void WaitPred(T* Futex, T ComparisonValue) {
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  T Result = AtomicFutex.load();
 
-  // Early exit if possible.
-  if (Result == ExpectedValue) {
-    return;
-  }
-
-  do {
+  while (!Pred {}(Result, ComparisonValue)) {
     Result = LoadExclusive(Futex);
-    if (Result == ExpectedValue) {
+    if (Pred {}(Result, ComparisonValue)) {
       return;
     }
-    Result = WFELoadAtomic(Futex);
-  } while (Result != ExpectedValue);
-}
 
-template void Wait<uint8_t>(uint8_t*, uint8_t);
-template void Wait<uint16_t>(uint16_t*, uint16_t);
-template void Wait<uint32_t>(uint32_t*, uint32_t);
-template void Wait<uint64_t>(uint64_t*, uint64_t);
+    Result = WFELoadAtomic(Futex);
+  }
+}
 
 template<typename T, typename TT>
 static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanoseconds& Timeout) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
 
-  T Result = AtomicFutex->load();
+  T Result = AtomicFutex.load();
 
   // Early exit if possible.
   if (Result == ExpectedValue) {
@@ -184,27 +180,43 @@ template bool Wait<uint16_t>(uint16_t*, uint16_t, const std::chrono::nanoseconds
 template bool Wait<uint32_t>(uint32_t*, uint32_t, const std::chrono::nanoseconds&);
 template bool Wait<uint64_t>(uint64_t*, uint64_t, const std::chrono::nanoseconds&);
 
-#else
-template<typename T, typename TT>
-static inline void Wait(T* Futex, TT ExpectedValue) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
-  T Result = AtomicFutex->load();
+template<typename T>
+static inline T OneShotWFEBitComparison(T* Futex, T Mask, T Comp) {
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  T Result = AtomicFutex.load();
 
   // Early exit if possible.
-  if (Result == ExpectedValue) {
-    return;
+  if ((Result & Mask) == Comp) {
+    return Result;
   }
 
-  do {
-    Result = AtomicFutex->load();
-  } while (Result != ExpectedValue);
+  Result = LoadExclusive(Futex);
+  if ((Result & Mask) == Comp) {
+    return Result;
+  }
+
+  // Waits for write and returns result.
+  Result = WFELoadAtomic(Futex);
+  return Result;
+}
+
+#else
+
+template<typename Pred, typename T>
+static inline void WaitPred(T* Futex, T ComparisonValue) {
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  T Result = AtomicFutex.load();
+
+  while (!Pred {}(Result, ComparisonValue)) {
+    Result = AtomicFutex.load();
+  }
 }
 
 template<typename T, typename TT>
 static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanoseconds& Timeout) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
 
-  T Result = AtomicFutex->load();
+  T Result = AtomicFutex.load();
 
   // Early exit if possible.
   if (Result == ExpectedValue) {
@@ -214,7 +226,7 @@ static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanosecon
   const auto Begin = std::chrono::high_resolution_clock::now();
 
   do {
-    Result = AtomicFutex->load();
+    Result = AtomicFutex.load();
 
     const auto CurrentCycleCounter = std::chrono::high_resolution_clock::now();
     if ((CurrentCycleCounter - Begin) >= Timeout) {
@@ -228,14 +240,24 @@ static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanosecon
 }
 #endif
 
+template<typename T, typename TT = T>
+static inline void Wait(T* Futex, TT ExpectedValue) {
+  WaitPred<std::equal_to<>, T>(Futex, ExpectedValue);
+}
+
+template void Wait<uint8_t>(uint8_t*, uint8_t);
+template void Wait<uint16_t>(uint16_t*, uint16_t);
+template void Wait<uint32_t>(uint32_t*, uint32_t);
+template void Wait<uint64_t>(uint64_t*, uint64_t);
+
 template<typename T>
 static inline void lock(T* Futex) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
   T Expected {};
   T Desired {1};
 
   // Try to CAS immediately.
-  if (AtomicFutex->compare_exchange_strong(Expected, Desired)) {
+  if (AtomicFutex.compare_exchange_strong(Expected, Desired)) {
     return;
   }
 
@@ -243,17 +265,17 @@ static inline void lock(T* Futex) {
     // Wait until the futex is unlocked.
     Wait(Futex, 0);
     Expected = 0;
-  } while (!AtomicFutex->compare_exchange_strong(Expected, Desired));
+  } while (!AtomicFutex.compare_exchange_strong(Expected, Desired));
 }
 
 template<typename T>
 static inline bool try_lock(T* Futex) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
   T Expected {};
   T Desired {1};
 
   // Try to CAS immediately.
-  if (AtomicFutex->compare_exchange_strong(Expected, Desired)) {
+  if (AtomicFutex.compare_exchange_strong(Expected, Desired)) {
     return true;
   }
 
@@ -262,8 +284,8 @@ static inline bool try_lock(T* Futex) {
 
 template<typename T>
 static inline void unlock(T* Futex) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
-  AtomicFutex->store(0);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  AtomicFutex.store(0);
 }
 
 #undef SPINLOOP_8BIT

FEXCore/Source/Utils/WritePriorityMutex.h

@@ -0,0 +1,381 @@
+// SPDX-License-Identifier: MIT
+#pragma once
+#include <atomic>
+#include <cstdint>
+
+#if !defined(_WIN32)
+#include <linux/futex.h> /* Definition of FUTEX_* constants */
+#include <sys/syscall.h> /* Definition of SYS_* constants */
+#include <unistd.h>
+#else
+#include <synchapi.h>
+#endif
+
+#include <FEXCore/Utils/LogManager.h>
+
+#include "Utils/SpinWaitLock.h"
+
+namespace FEXCore::Utils::WritePriorityMutex {
+
+// A custom mutex that prioritizes exclusive locks.
+// In highly contested scenarios, this can help minimize overall contention time.
+//
+// Features:
+//  - Up to 32767 pending exclusive locks ("writers")
+//  - Up to 32767 pending shared_locks ("readers")
+//  - Low-overhead waiting via WFE with a fallback to futex on timeout
+//  - Direct writer->reader hand-off and vice-versa to further reduce overhead
+//
+// Trade-offs:
+//  - No guaranteed order of wake-ups besides prioritizing writers
+//  - No support for recursive locking
+//  - We can't use FUTEX_LOCK_PI to enable priority inheritance
+class Mutex final {
+public:
+  Mutex() = default;
+
+  // Move-only type
+  Mutex(const Mutex&) = delete;
+  Mutex& operator=(const Mutex&) = delete;
+  Mutex(Mutex&& rhs) = delete;
+  Mutex& operator=(Mutex&&) = delete;
+
+  void lock() {
+    // Try a non-blocking lock first.
+    if (try_lock()) {
+      return;
+    }
+
+    // Try a quick WFE write-lock.
+    if (Attempt_WFE_WriteLock()) {
+      return;
+    }
+
+    // Still couldn't get it. Start waiting.
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Expected {};
+    uint32_t Desired {};
+#if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
+    Expected = AtomicFutex.load(std::memory_order_relaxed);
+    do {
+      // Increment the number of write waiters.
+      Desired = Expected + WRITE_WAITER_INCREMENT;
+
+      LOGMAN_THROW_A_FMT((Desired & WRITE_WAITER_COUNT_MASK) != 0, "Overflow in write-waiters!");
+    } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+#else
+    Expected = AtomicFutex.fetch_add(WRITE_WAITER_INCREMENT);
+    Desired = Expected + WRITE_WAITER_INCREMENT;
+#endif
+
+    // Thread added to waiter list.
+    Expected = Desired;
+
+    while (true) {
+      bool Sleep = false;
+
+      do {
+        if ((Expected & WRITE_OWNED_BIT) == 0 && (Expected & READ_OWNER_COUNT_MASK) == 0) {
+          // If not write-owned, and no read-owners, try to acquire.
+          LOGMAN_THROW_A_FMT((Expected & WRITE_WAITER_COUNT_MASK) != 0, "Underflow in write-waiters!");
+
+          // Add write-owned bit.
+          Desired = Expected | WRITE_OWNED_BIT;
+
+          // Remove ourselves from the wait list.
+          Desired -= WRITE_WAITER_INCREMENT;
+
+          Sleep = false;
+        } else {
+          // Already write-owned or read-locked. Go to sleep.
+          Desired = Expected;
+          Sleep = true;
+          break;
+        }
+
+      } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+
+      if (!Sleep) {
+        // Acquired early.
+        LOGMAN_THROW_A_FMT((Desired & WRITE_OWNED_BIT) == WRITE_OWNED_BIT, "Somehow acquired a write-lock without it being set!");
+        return;
+      }
+      FutexWaitForWriteAvailable(Desired);
+
+      Expected = AtomicFutex.load(std::memory_order_relaxed);
+    }
+  }
+
+  void lock_shared() {
+    // Try an uncontended lock first.
+    if (try_lock_shared()) {
+      return;
+    }
+
+    // Try a quick WFE read-lock.
+    if (Attempt_WFE_ReadLock()) {
+      return;
+    }
+
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+    uint32_t Desired {};
+
+    while (true) {
+      bool Sleep = false;
+      do {
+        if ((Expected & WRITE_OWNED_BIT) == 0 && (Expected & WRITE_WAITER_COUNT_MASK) == 0) {
+          // If no write-owner and no write-waiting, try and acquire.
+
+          Desired = Expected + READ_OWNER_INCREMENT;
+          LOGMAN_THROW_A_FMT((Desired & READ_OWNER_COUNT_MASK) != 0, "Overflow in read-owners!");
+          Sleep = false;
+        } else {
+          // Waiting for lock to become available. Add to waiters.
+          Desired = Expected | READ_WAITER_BIT;
+          Sleep = true;
+        }
+      } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+
+      if (!Sleep) {
+        // Acquired early.
+        LOGMAN_THROW_A_FMT((Desired & WRITE_OWNED_BIT) != WRITE_OWNED_BIT, "Somehow read-locked and got a write lock!");
+        return;
+      }
+
+      FutexWaitForReadAvailable(Desired);
+
+      Expected = AtomicFutex.load(std::memory_order_relaxed);
+    }
+  }
+
+  void unlock() {
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+    uint32_t Desired {};
+    do {
+      LOGMAN_THROW_A_FMT((Expected & WRITE_OWNED_BIT) == WRITE_OWNED_BIT, "Trying to write-unlock something not write-locked!");
+      // Remove the exclusive lock bit.
+      Desired = Expected & ~WRITE_OWNED_BIT;
+
+      // If no more writers, then make sure to clear the read-waiters bit as well.
+      if ((Desired & WRITE_WAITER_COUNT_MASK) == 0) {
+        Desired &= ~READ_WAITER_BIT;
+      }
+    } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+
+    // If success, then `Expected` has old value. Containing `READ_WAITER_BIT` which was just masked off, and also `WRITE_WAITER_COUNT_MASK`.
+    if ((Expected & WRITE_WAITER_COUNT_MASK)) {
+      // Handle write-write handoff.
+      FutexWakeWriter();
+    } else if ((Expected & READ_WAITER_BIT)) {
+      // Handle write-reader handoff.
+      FutexWakeReaders();
+    }
+  }
+
+  void unlock_shared() {
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Desired {};
+#if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+    do {
+      LOGMAN_THROW_A_FMT((Expected & WRITE_OWNED_BIT) != WRITE_OWNED_BIT, "Trying to read-unlock something write-locked!");
+      LOGMAN_THROW_A_FMT((Expected & READ_OWNER_COUNT_MASK) != 0, "Trying to read-unlock something not read-locked!");
+
+      // Decrement the shared counter.
+      Desired = Expected - READ_OWNER_INCREMENT;
+    } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+#else
+    Desired = AtomicFutex.fetch_sub(READ_OWNER_INCREMENT) - READ_OWNER_INCREMENT;
+#endif
+
+    // Handle read->write handoff if there are any waiting writers, and no readers left.
+    if ((Desired & WRITE_WAITER_COUNT_MASK) && (Desired & READ_OWNER_COUNT_MASK) == 0) {
+      FutexWakeWriter();
+    }
+  }
+
+  bool try_lock() {
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Expected = 0;
+
+    // Try and grab the owned bit.
+    uint32_t Desired = WRITE_OWNED_BIT;
+
+    // try to CAS immediately.
+    return AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire);
+  }
+
+  // Can race with other threads trying to lock shared!
+  bool try_lock_shared() {
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+
+    // Exclusively owned or has a list of waiting owners. Can't pass.
+    if ((Expected & WRITE_OWNED_BIT) || (Expected & WRITE_WAITER_COUNT_MASK)) {
+      return false;
+    }
+
+    // Try to add reader.
+    uint32_t Desired = Expected + READ_OWNER_INCREMENT;
+    LOGMAN_THROW_A_FMT((Desired & READ_OWNER_COUNT_MASK) != 0, "Overflow in read-owners!");
+
+    // Uncontended mutex check
+    return AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire);
+  }
+
+#if !defined(_WIN32)
+  // Initialize the internal mutex object to its default initializer state.
+  // Should only ever be used in the child process when a Linux fork() has occured.
+  void StealAndDropActiveLocks() {
+    Futex = 0;
+  }
+#endif
+
+private:
+
+#if !defined(_WIN32)
+  void FutexWaitForWriteAvailable(uint32_t Expected) {
+    ::syscall(SYS_futex, &Futex, FUTEX_PRIVATE_FLAG | FUTEX_WAIT_BITSET, Expected, nullptr, nullptr, FUTEX_BITSET_WAIT_WRITERS);
+  }
+
+  // Read-lock waiting for writers to drain out.
+  void FutexWaitForReadAvailable(uint32_t Expected) {
+    ::syscall(SYS_futex, &Futex, FUTEX_PRIVATE_FLAG | FUTEX_WAIT_BITSET, Expected, nullptr, nullptr, FUTEX_BITSET_WAIT_READERS);
+  }
+
+  // Read-Lock or Write-lock unlocked, wake one writer.
+  // - Read->Write handoff.
+  // - Write->Write handoff.
+  void FutexWakeWriter() {
+    ::syscall(SYS_futex, &Futex, FUTEX_PRIVATE_FLAG | FUTEX_WAKE_BITSET, 1, nullptr, nullptr, FUTEX_BITSET_WAIT_WRITERS);
+  }
+
+  // Write-lock unlocked, wake read-locks waiting.
+  void FutexWakeReaders() {
+    // Wake all readers.
+    ::syscall(SYS_futex, &Futex, FUTEX_PRIVATE_FLAG | FUTEX_WAKE_BITSET, INT_MAX, nullptr, nullptr, FUTEX_BITSET_WAIT_READERS);
+  }
+#else
+  // Writers wait for the full 32-bit futex.
+  void FutexWaitForWriteAvailable(uint32_t Expected) {
+    WaitOnAddress(&Futex, &Expected, sizeof(Futex), INFINITE);
+  }
+
+  // Readers wait for Futex bits [31:16] to be zero.
+  void FutexWaitForReadAvailable(uint32_t Expected) {
+    auto ReadWaiterAddress = reinterpret_cast<uint8_t*>(&Futex) + 2;
+    uint16_t smol_Expected = Expected >> 16;
+    WaitOnAddress(ReadWaiterAddress, &smol_Expected, sizeof(smol_Expected), INFINITE);
+  }
+
+  void FutexWakeWriter() {
+    WakeByAddressSingle(&Futex);
+  }
+
+  void FutexWakeReaders() {
+    auto ReadWaiterAddress = reinterpret_cast<uint8_t*>(&Futex) + 2;
+    WakeByAddressAll(ReadWaiterAddress);
+  }
+#endif
+
+  // Reuse the SpinWaitLock WFE implementations for read/write lock acquiring with WFE.
+  // Can't reuse the spin-lock directly as some bit-representations are different.
+  // WFE-write-lock is less likely to occur the more read-lock threads are participating. Can still occur so good to try.
+  // WFE-read-lock is actually quite likely to succeed.
+  // Return: true if the lock was acquired.
+  bool Attempt_WFE_WriteLock() {
+#ifdef _M_ARM_64
+    const auto Begin = FEXCore::Utils::SpinWaitLock::GetCycleCounter();
+    auto Now = Begin;
+    const auto Duration = FEXCore::Utils::SpinWaitLock::CycleCounterFrequency / CYCLECOUNT_DIVISOR;
+
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+
+    while ((Now - Begin) < Duration) {
+      if (Expected == 0) {
+        // Try and grab the owned bit.
+        uint32_t Desired = WRITE_OWNED_BIT;
+
+        if (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire)) {
+          return true;
+        }
+      }
+
+      // One-shot attempt to wait for mask to be zero.
+      Expected = FEXCore::Utils::SpinWaitLock::OneShotWFEBitComparison(&Futex, ~0U, 0U);
+      Now = FEXCore::Utils::SpinWaitLock::GetCycleCounter();
+    }
+#endif
+
+    return false;
+  }
+
+  // Return: true if the lock was acquired.
+  bool Attempt_WFE_ReadLock() {
+#ifdef _M_ARM_64
+    // Spin on a WFE for a short-amount of time, waiting for write-owned and writer-count to be zero.
+    //  - Attempt to acquire read-lock at that point.
+    //  - Don't add read-waiters bit on failure, return false.
+    const auto Begin = FEXCore::Utils::SpinWaitLock::GetCycleCounter();
+    auto Now = Begin;
+    const auto Duration = FEXCore::Utils::SpinWaitLock::CycleCounterFrequency / CYCLECOUNT_DIVISOR;
+
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+    uint32_t Desired {};
+
+    while ((Now - Begin) < Duration) {
+      if ((Expected & WRITE_OWNED_BIT) == 0 && (Expected & WRITE_WAITER_COUNT_MASK) == 0) {
+        // If no write-owner and no write-waiting, try and acquire.
+
+        Desired = Expected + READ_OWNER_INCREMENT;
+        LOGMAN_THROW_A_FMT((Desired & READ_OWNER_COUNT_MASK) != 0, "Overflow in read-owners!");
+        if (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire)) {
+          return true;
+        }
+      }
+
+      // One-shot attempt to wait for mask to be zero.
+      Expected = FEXCore::Utils::SpinWaitLock::OneShotWFEBitComparison(&Futex, WRITE_OWNED_BIT | WRITE_WAITER_COUNT_MASK, 0U);
+      Now = FEXCore::Utils::SpinWaitLock::GetCycleCounter();
+    }
+#endif
+
+    return false;
+  }
+
+  constexpr static uint32_t WRITE_OWNED_BIT = 1U << 31;
+  constexpr static uint32_t READ_WAITER_BIT = 1U << 15;
+  constexpr static uint32_t WRITE_WAITER_OFFSET = 16;
+  constexpr static uint32_t WRITE_WAITER_INCREMENT = 1U << WRITE_WAITER_OFFSET;
+  constexpr static uint32_t READ_OWNER_INCREMENT = 1;
+
+  // Count masks
+  constexpr static uint32_t WRITE_WAITER_COUNT_MASK = 0x7FFFU << WRITE_WAITER_OFFSET;
+  constexpr static uint32_t READ_OWNER_COUNT_MASK = 0x7FFFU;
+
+  // Independent futex bit-set masks.
+  // Wait for readers to drain.
+  constexpr static uint32_t FUTEX_BITSET_WAIT_READERS = 1U << 0;
+  // Wait for writers to drain.
+  constexpr static uint32_t FUTEX_BITSET_WAIT_WRITERS = 1U << 1;
+
+  // Only spin on WFE for 0.01ms (10k ns).
+  constexpr static uint64_t CYCLECOUNT_DIVISOR = 1'000'000'000ULL / 10'000U;
+
+  // Layout:
+  //    Bits[31]: Write-lock bit.
+  // Bits[30:16]: Write-waiter count.
+  //    Bits[15]: Read-waiter bit.
+  //  Bits[14:0]: Read-owner count.
+  uint32_t Futex {};
+};
+} // namespace FEXCore::Utils::WritePriorityMutex

FEXCore/include/FEXCore/Config/Config.h

@@ -103,28 +103,25 @@ static inline std::optional<fextl::string> EnumParser(const ArrayPairType& EnumP
   return fextl::fmt::format("{}", EnumMask);
 }
 
-namespace DefaultValues {
-#define P(x) x
-#define OPT_BASE(type, group, enum, json, default) extern const P(type) P(enum);
-#define OPT_STR(group, enum, json, default) extern const std::string_view P(enum);
-#define OPT_STRARRAY(group, enum, json, default) OPT_STR(group, enum, json, default)
-#include <FEXCore/Config/ConfigValues.inl>
+using StringArrayType = fextl::list<fextl::string>;
 
-  namespace Type {
-    using StringArrayType = fextl::list<fextl::string>;
-#define OPT_BASE(type, group, enum, json, default) using P(enum) = P(type);
-#define OPT_STR(group, enum, json, default) using P(enum) = fextl::string;
-#define OPT_STRARRAY(group, enum, json, default) using P(enum) = StringArrayType;
+namespace detail {
+  template<ConfigOption Option>
+  struct ConfigOptionInfo;
+#define DEFINE_METAINFO(type, enum, default)             \
+  template<>                                             \
+  struct ConfigOptionInfo<ConfigOption::CONFIG_##enum> { \
+    using Type = type;                                   \
+    static auto Default() {                              \
+      extern default;                                    \
+      return enum;                                       \
+    }                                                    \
+  };
+#define OPT_BASE(type, group, enum, json, default) DEFINE_METAINFO(type, enum, const type enum)
+#define OPT_STR(group, enum, json, default) DEFINE_METAINFO(fextl::string, enum, const std::string_view enum)
+#define OPT_STRARRAY(group, enum, json, default) DEFINE_METAINFO(StringArrayType, enum, const std::string_view enum)
 #include <FEXCore/Config/ConfigValues.inl>
-  } // namespace Type
-#define FEX_CONFIG_OPT(name, enum)                                          \
-  FEXCore::Config::Value<FEXCore::Config::DefaultValues::Type::enum> name { \
-    FEXCore::Config::CONFIG_##enum,                                         \
-    FEXCore::Config::DefaultValues::enum                                    \
-  }
-
-#undef P
-} // namespace DefaultValues
+} // namespace detail
 
 FEX_DEFAULT_VISIBILITY void SetDataDirectory(std::string_view Path, bool Global);
 FEX_DEFAULT_VISIBILITY void SetConfigDirectory(const std::string_view Path, bool Global);
@@ -135,8 +132,7 @@ FEX_DEFAULT_VISIBILITY const fextl::string& GetConfigDirectory(bool Global);
 FEX_DEFAULT_VISIBILITY const fextl::string& GetConfigFileLocation(bool Global = false);
 FEX_DEFAULT_VISIBILITY fextl::string GetApplicationConfig(const std::string_view Program, bool Global);
 
-using LayerValue =
-  std::variant< fextl::string, DefaultValues::Type::StringArrayType, uint8_t, int8_t, uint16_t, int16_t, uint32_t, int32_t, uint64_t, int64_t, bool >;
+using LayerValue = std::variant< fextl::string, StringArrayType, uint8_t, int8_t, uint16_t, int16_t, uint32_t, int32_t, uint64_t, int64_t, bool >;
 
 using LayerOptions = fextl::unordered_map<ConfigOption, LayerValue>;
 
@@ -151,16 +147,16 @@ public:
     return OptionMap.find(Option) != OptionMap.end();
   }
 
-  std::optional<DefaultValues::Type::StringArrayType*> All(ConfigOption Option) {
+  std::optional<StringArrayType*> All(ConfigOption Option) {
     const auto it = OptionMap.find(Option);
     if (it == OptionMap.end()) {
       return std::nullopt;
     }
 
     auto& Value = it->second;
-    LOGMAN_THROW_A_FMT(std::holds_alternative<DefaultValues::Type::StringArrayType>(Value), "Tried to get config of invalid type!");
+    LOGMAN_THROW_A_FMT(std::holds_alternative<StringArrayType>(Value), "Tried to get config of invalid type!");
 
-    return &std::get<DefaultValues::Type::StringArrayType>(Value);
+    return &std::get<StringArrayType>(Value);
   }
 
   std::optional<fextl::string*> Get(ConfigOption Option) {
@@ -201,12 +197,12 @@ public:
     auto it = OptionMap.find(Option);
     if (it == OptionMap.end()) {
       // If the option didn't exist as a StringArrayType yet, emplace it.
-      it = OptionMap.emplace(Option, DefaultValues::Type::StringArrayType {}).first;
+      it = OptionMap.emplace(Option, StringArrayType {}).first;
     }
 
     auto& Value = it->second;
-    LOGMAN_THROW_A_FMT(std::holds_alternative<DefaultValues::Type::StringArrayType>(Value), "Tried to get config of invalid type!");
-    std::get<DefaultValues::Type::StringArrayType>(Value).emplace_back(Data);
+    LOGMAN_THROW_A_FMT(std::holds_alternative<StringArrayType>(Value), "Tried to get config of invalid type!");
+    std::get<StringArrayType>(Value).emplace_back(Data);
   }
 
   void Erase(ConfigOption Option) {
@@ -236,7 +232,9 @@ FEX_DEFAULT_VISIBILITY fextl::string FindContainerPrefix();
 FEX_DEFAULT_VISIBILITY void AddLayer(fextl::unique_ptr<FEXCore::Config::Layer> _Layer);
 
 FEX_DEFAULT_VISIBILITY bool Exists(ConfigOption Option);
-FEX_DEFAULT_VISIBILITY std::optional<DefaultValues::Type::StringArrayType*> All(ConfigOption Option);
+FEX_DEFAULT_VISIBILITY std::optional<StringArrayType*> All(ConfigOption Option);
+template<typename T>
+FEX_DEFAULT_VISIBILITY std::optional<T> GetConv(ConfigOption Option);
 FEX_DEFAULT_VISIBILITY std::optional<fextl::string*> Get(ConfigOption Option);
 FEX_DEFAULT_VISIBILITY void Set(ConfigOption Option, std::string_view Data);
 FEX_DEFAULT_VISIBILITY void Erase(ConfigOption Option);
@@ -271,18 +269,18 @@ public:
     return ValueData;
   }
 
-  Value(T Value) requires (!std::is_same_v<T, DefaultValues::Type::StringArrayType>)
+  Value(T Value) requires (!std::is_same_v<T, StringArrayType>)
   {
     ValueData = std::move(Value);
   }
 
   // Array value types.
-  Value(FEXCore::Config::ConfigOption Option, std::string_view) requires (std::is_same_v<T, DefaultValues::Type::StringArrayType>)
+  Value(FEXCore::Config::ConfigOption Option, std::string_view) requires (std::is_same_v<T, StringArrayType>)
   {
     GetListIfExists(Option, &ValueData);
   }
 
-  DefaultValues::Type::StringArrayType& All() requires (std::is_same_v<T, DefaultValues::Type::StringArrayType>)
+  StringArrayType& All() requires (std::is_same_v<T, StringArrayType>)
   {
     return ValueData;
   }
@@ -293,6 +291,38 @@ private:
   static T GetIfExists(FEXCore::Config::ConfigOption Option, T Default);
   static T GetIfExists(FEXCore::Config::ConfigOption Option, std::string_view Default);
 
-  static void GetListIfExists(FEXCore::Config::ConfigOption Option, DefaultValues::Type::StringArrayType* List);
+  static void GetListIfExists(FEXCore::Config::ConfigOption Option, StringArrayType* List);
 };
+
+/**
+ * Wrapper around Value that automatically picks the default for the given ConfigOption
+ */
+template<ConfigOption Option>
+struct FEX_DEFAULT_VISIBILITY Getter : public Value<typename detail::ConfigOptionInfo<Option>::Type> {
+  using OptionInfo = detail::ConfigOptionInfo<Option>;
+  Getter()
+    : Value<typename OptionInfo::Type> {Option, OptionInfo::Default()} {}
+};
+
+/**
+ * Helper for reading a config value with caching.
+ *
+ * Typically this is used to declare class members so that the value is read
+ * on construction of the parent.
+ */
+#define FEX_CONFIG_OPT(name, enum) FEXCore::Config::Getter<FEXCore::Config::ConfigOption::CONFIG_##enum> name {}
+
+#define OPT_BASE(type, group, enum, json, default)                              \
+  /**                                                                           \
+   *  Helper for reading a config value.                                        \
+   *                                                                            \
+   *  In contrast to FEX_CONFIG_OPT, this can be used in arbitrary expressions, \
+   *  at the expense of not caching the value. Use Getter instead if the value  \
+   *  is read frequently.                                                       \
+   */                                                                           \
+  inline auto Get_##enum() {                                                    \
+    return Getter<FEXCore::Config::ConfigOption::CONFIG_##enum> {};             \
+  }
+#include <FEXCore/Config/ConfigValues.inl>
+
 } // namespace FEXCore::Config

FEXCore/include/FEXCore/Core/CodeCache.h

@@ -1,10 +1,20 @@
 // SPDX-License-Identifier: MIT
 #pragma once
 
+#include <FEXCore/fextl/functional.h>
+#include <FEXCore/fextl/map.h>
 #include <FEXCore/fextl/memory.h>
+#include <FEXCore/fextl/set.h>
 #include <FEXCore/fextl/string.h>
+#include <FEXCore/fextl/vector.h>
 
+#include <atomic>
 #include <cstdint>
+#include <mutex>
+#include <optional>
+#include <shared_mutex>
+#include <span>
+#include <unistd.h>
 
 namespace FEXCore {
 
@@ -20,8 +30,14 @@ namespace HLE {
 struct ExecutableFileInfo {
   ~ExecutableFileInfo();
 
+#if __clang_major__ < 16
+  // Workaround for broken aggregate-initialization with std::piecewise_construct
+  ExecutableFileInfo(fextl::unique_ptr<HLE::SourcecodeMap>, uint64_t, fextl::string);
+  ExecutableFileInfo() = default;
+#endif
+
   fextl::unique_ptr<HLE::SourcecodeMap> SourcecodeMap;
-  fextl::string FileId;
+  uint64_t FileId = 0;
   fextl::string Filename;
 };
 
@@ -33,10 +49,129 @@ struct ExecutableFileSectionInfo {
   uintptr_t FileStartVA;
 };
 
+using CodeMapFileId = uint64_t;
+
+/**
+ * Code maps capture information required for offline code cache generation
+ * and are written to disk during execution of FEX.
+ *
+ * Almost all CodeMap data will be an Entry that indicates blocks to be
+ * compiled for cache generation. The reserved value `LoadExternalLibrary`
+ * indicates that an instance of ExternalLibraryInfo follows (the entry data
+ * itself should be skipped in that case).
+ */
+struct CodeMap {
+  // Describes the location of an entry block compiled during execution
+  struct FEX_PACKED Entry {
+    CodeMapFileId FileId;
+    uint32_t BlockOffset;
+  };
+
+  // Describes an external library referenced during execution
+  struct ExternalLibraryInfo {
+    CodeMapFileId ExternalFileId;
+
+    // null-terminated file path; EITHER relative to the main executable OR an absolute path OR starting with a magic identifier:
+    // - WINE/: Path to Wine/Proton installation
+    // - WINEPREFIX/: Path to Wine/Proton prefix
+    // - SLR/: Path to Steam Linux Runtime
+    // At runtime, FEX will always dump absolute paths
+    char Path[];
+    // Followed by padding to a 4 byte boundary
+  };
+
+  // Followed by ExternalLibraryInfo
+  static constexpr Entry LoadExternalLibrary = {0xffff'ffff'ffff'ffff, 0xffff'ffff};
+
+  struct FEX_PACKED SetExecutableFileId {
+    Entry Marker = {0xffff'ffff'ffff'ffff, 0xffff'fffe};
+    CodeMapFileId ExecutableFileId;
+  };
+
+  struct ParsedContents {
+    fextl::string Filename;
+    fextl::set<uint64_t> Blocks;
+    bool IsExecutable = false;
+  };
+
+  // Follows scheme fileid[-nomb]
+  // The nomb ("no multiblock") suffix signifies that the code map is for use without multiblock, only.
+  static fextl::string GetBaseFilename(const ExecutableFileInfo& MainExecutable, bool AddNombSuffix);
+
+  static fextl::map<CodeMapFileId, ParsedContents> ParseCodeMap(std::ifstream& File);
+};
+
+struct CodeMapOpener {
+  virtual ~CodeMapOpener() = default;
+  virtual int OpenCodeMapFile() = 0;
+};
+
+class CodeMapWriter {
+public:
+  CodeMapWriter(CodeMapOpener&, bool OpenEagerly = false);
+  ~CodeMapWriter();
+
+  // Checks if writing is enabled. Calls to this functions may also be interpreted as signals that writes are about to happen
+  bool IsWriteEnabled(const ExecutableFileSectionInfo&);
+
+  void ResetAfterFork() {
+    if (CodeMapFD.value_or(-1) != -1) {
+      close(CodeMapFD.value());
+      CodeMapFD.reset();
+    }
+    BufferOffset = 0;
+    KnownFileIds.clear();
+  }
+
+  bool IsBackingFD(int FD) const {
+    if (FD == CodeMapFD) {
+      LogMan::Msg::DFmt("Hiding directory entry for code map FD");
+      return true;
+    }
+    return false;
+  }
+
+  void AppendBlock(const FEXCore::ExecutableFileSectionInfo&, uint64_t Entry);
+  void AppendLibraryLoad(const FEXCore::ExecutableFileInfo&);
+  void AppendSetMainExecutable(const FEXCore::ExecutableFileInfo&);
+
+  // Thread-safely commit any pending data to disk
+  void Flush(size_t Offset);
+
+private:
+  // Queues data into an internal ring buffer.
+  // Call Flush() to commit the data to disk.
+  void AppendData(std::span<const std::byte> Data);
+
+  // Commit given data range to disk
+  void Flush(size_t Offset, std::unique_lock<std::shared_mutex>&);
+
+  std::shared_mutex Mutex;
+  fextl::vector<std::byte> Buffer;
+  std::atomic<size_t> BufferOffset {0};
+
+  fextl::set<CodeMapFileId> KnownFileIds;
+
+  // std::nullopt: We haven't requested a CodeMapFD yet
+  // value is -1:  We requested a CodeMapFD but FEXServer told us not to write any data
+  // other values: Code map writing is active
+  std::optional<int> CodeMapFD;
+
+  CodeMapOpener& FileOpener;
+};
+
 class AbstractCodeCache {
 public:
   virtual ~AbstractCodeCache() = default;
 
+  /**
+   * Computes a unique identifier for the referenced binary file to be used for
+   * generating the code map.
+   * This identifier is independent of FEX build/runtime configuration and
+   * stable across FEX updates.
+   */
+  virtual uint64_t ComputeCodeMapId(std::string_view Filename, int FD) = 0;
+
   /**
    * Loads a code cache from mapped memory and appends it to the current Core state.
    * TODO: Optionally recompiles all contained code blocks at runtime for validation.

FEXCore/include/FEXCore/Core/Context.h

@@ -42,9 +42,6 @@ enum OperatingMode {
 
 using CodeRangeInvalidationFn = std::function<void(uint64_t start, uint64_t Length)>;
 
-// Nested vector of guest block entrypoints
-using InvalidatedEntryAccumulator = fextl::vector<fextl::vector<uint64_t>>;
-
 using CustomIREntrypointHandler = std::function<void(uintptr_t Entrypoint, IR::IREmitter*)>;
 
 using ExitHandler = std::function<void(Core::InternalThreadState* Thread)>;
@@ -139,10 +136,13 @@ public:
   FEX_DEFAULT_VISIBILITY virtual FEXCore::CPUID::FunctionResults RunCPUIDFunctionName(uint32_t Function, uint32_t Leaf, uint32_t CPU) = 0;
 
   virtual AbstractCodeCache& GetCodeCache() = 0;
+  virtual void SetCodeMapWriter(fextl::unique_ptr<CodeMapWriter>) = 0;
+  virtual void FlushAndCloseCodeMap() = 0;
 
   FEX_DEFAULT_VISIBILITY virtual void ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, bool NewCodeBuffer = true) = 0;
-  FEX_DEFAULT_VISIBILITY virtual void InvalidateGuestCodeRange(
-    FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator, uint64_t Start, uint64_t Length) = 0;
+  FEX_DEFAULT_VISIBILITY virtual void InvalidateCodeBuffersCodeRange(uint64_t Start, uint64_t Length) = 0;
+  FEX_DEFAULT_VISIBILITY virtual void
+  InvalidateThreadCachedCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) = 0;
   FEX_DEFAULT_VISIBILITY virtual FEXCore::ForkableSharedMutex& GetCodeInvalidationMutex() = 0;
 
   FEX_DEFAULT_VISIBILITY virtual void

FEXCore/include/FEXCore/Core/CoreState.h

@@ -104,6 +104,10 @@ struct CPUState {
   uint64_t avx_high[16][2];
 
   uint64_t gregs[16] {};
+  uint64_t L1Pointer {};
+  uint64_t L1Mask {};
+  uint64_t callret_sp {};
+  uint64_t _pad1 {};
   XMMRegs xmm {};
 
   // Raw segment register indexes
@@ -116,8 +120,6 @@ struct CPUState {
   uint64_t gs_cached {};
   uint64_t fs_cached {};
   uint8_t flags[48] {};
-  uint64_t callret_sp {};
-  uint64_t _pad1 {};
   uint64_t mm[8][2] {};
 
   // 32bit x86 state
@@ -247,6 +249,8 @@ static_assert(offsetof(CPUState, xmm) % 32 == 0, "xmm needs to be 256-bit aligne
 static_assert(offsetof(CPUState, mm) % 16 == 0, "mm needs to be 128-bit aligned!");
 static_assert(offsetof(CPUState, gregs[15]) <= 504, "gregs maximum offset must be <= 504 for ldp/stp to work");
 static_assert(offsetof(CPUState, DeferredSignalRefCount) % 8 == 0, "Needs to be 8-byte aligned");
+static_assert(offsetof(CPUState, L1Pointer) <= 504, "This needs to be <= 504 for ldp");
+static_assert(offsetof(CPUState, L1Mask) == (offsetof(CPUState, L1Pointer) + 8), "These two variables are paired");
 
 struct InternalThreadState;
 
@@ -349,13 +353,11 @@ struct JITPointers {
     uint64_t ExitFunctionLinker {};
     uint64_t ThreadStopHandlerSpillSRA {};
     uint64_t ThreadPauseHandlerSpillSRA {};
-    uint64_t UnimplementedInstructionHandler {};
     uint64_t GuestSignal_SIGILL {};
     uint64_t GuestSignal_SIGTRAP {};
     uint64_t GuestSignal_SIGSEGV {};
     uint64_t SignalReturnHandler {};
     uint64_t SignalReturnHandlerRT {};
-    uint64_t L1Pointer {};
     uint64_t L2Pointer {};
     /**  @} */
 
@@ -368,8 +370,6 @@ struct JITPointers {
       // Process specific
       uint64_t LUDIV {};
       uint64_t LDIV {};
-      uint64_t LUREM {};
-      uint64_t LREM {};
 
       // Thread Specific
 
@@ -378,8 +378,6 @@ struct JITPointers {
        * @{ */
       uint64_t LUDIVHandler {};
       uint64_t LDIVHandler {};
-      uint64_t LUREMHandler {};
-      uint64_t LREMHandler {};
       /**  @} */
     } AArch64;
 

FEXCore/include/FEXCore/Core/HostFeatures.h

@@ -40,6 +40,7 @@ struct HostFeatures {
   bool SupportsECV {};
   bool SupportsWFXT {};
   bool Supports3DNow {};
+  bool SupportsSSE4a {};
 
   // Float exception behaviour
   bool SupportsAFP {};

FEXCore/include/FEXCore/Core/SignalDelegator.h

@@ -67,6 +67,10 @@ public:
     return Config;
   }
 
+  virtual uintptr_t GetThunkCallbackRET() const {
+    return 0;
+  }
+
 protected:
   SignalDelegatorConfig Config;
 };

FEXCore/include/FEXCore/Debug/InternalThreadState.h

@@ -4,6 +4,7 @@
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Utils/AllocatorHooks.h>
 #include <FEXCore/Utils/TypeDefines.h>
+#include <FEXCore/Utils/LongJump.h>
 #include <FEXCore/fextl/memory.h>
 #include <FEXCore/fextl/vector.h>
 
@@ -80,7 +81,14 @@ private:
 static_assert(!std::is_move_constructible_v<NonMovableUniquePtr<int>>);
 static_assert(!std::is_move_assignable_v<NonMovableUniquePtr<int>>);
 
-struct InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
+// Store used for unaligned LDAXR*/STLXR* emulation.
+struct UnalignedExclusiveStore {
+  uint64_t Addr;
+  uint64_t Store;
+  uint8_t Size;
+};
+
+struct alignas(FEXCore::Utils::FEX_PAGE_SIZE) InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
   FEXCore::Core::CpuStateFrame* const CurrentFrame = &BaseFrameState;
 
   FEXCore::Context::Context* const CTX;
@@ -101,6 +109,8 @@ struct InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
   // This pointer is owned by the frontend.
   FEXCore::SHMStats::ThreadStats* ThreadStats {};
 
+  UnalignedExclusiveStore ExclusiveStore;
+
   ///< Data pointer for exclusive use by the frontend
   void* FrontendPtr;
 
@@ -109,6 +119,10 @@ struct InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
   // The low address of the call-ret stack allocation (not including guard pages)
   void* CallRetStackBase {};
 
+  uintptr_t JITGuardPage {};
+  uint64_t JITGuardOverflowArgument {};
+  FEXCore::UncheckedLongJump::JumpBuf RestartJump;
+
   // BaseFrameState should always be at the end, directly before the interrupt fault page
   alignas(16) FEXCore::Core::CpuStateFrame BaseFrameState {};
 
@@ -116,8 +130,9 @@ struct InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
   alignas(FEXCore::Utils::FEX_PAGE_SIZE) uint8_t InterruptFaultPage[FEXCore::Utils::FEX_PAGE_SIZE];
 };
 static_assert(std::is_standard_layout_v<FEXCore::Core::InternalThreadState>);
-static_assert(
-  (offsetof(FEXCore::Core::InternalThreadState, InterruptFaultPage) - offsetof(FEXCore::Core::InternalThreadState, BaseFrameState)) < 4096,
-  "Fault page is outside of immediate range from CPU state");
+static_assert((offsetof(FEXCore::Core::InternalThreadState, InterruptFaultPage) - offsetof(FEXCore::Core::InternalThreadState, BaseFrameState)) <
+                FEXCore::Utils::FEX_PAGE_SIZE,
+              "Fault page is outside of immediate range from CPU state");
+static_assert(sizeof(FEXCore::Core::InternalThreadState) == (FEXCore::Utils::FEX_PAGE_SIZE * 2));
 
 } // namespace FEXCore::Core

FEXCore/include/FEXCore/HLE/SourcecodeResolver.h

@@ -84,6 +84,6 @@ private:
 
 class SourcecodeResolver {
 public:
-  virtual fextl::unique_ptr<SourcecodeMap> GenerateMap(const std::string_view& GuestBinaryFile, const std::string_view& GuestBinaryFileId) = 0;
+  virtual fextl::unique_ptr<SourcecodeMap> GenerateMap(std::string_view GuestBinaryFile, std::string_view GuestBinaryFileId) = 0;
 };
 } // namespace FEXCore::HLE

FEXCore/include/FEXCore/HLE/SyscallHandler.h

@@ -54,10 +54,6 @@ public:
   virtual ~SyscallHandler() = default;
 
   virtual uint64_t HandleSyscall(FEXCore::Core::CpuStateFrame* Frame, FEXCore::HLE::SyscallArguments* Args) = 0;
-  virtual SyscallABI GetSyscallABI(uint64_t Syscall) = 0;
-  virtual FEXCore::IR::SyscallFlags GetSyscallFlags(uint64_t Syscall) const {
-    return FEXCore::IR::SyscallFlags::DEFAULT;
-  }
 
   SyscallOSABI GetOSABI() const {
     return OSABI;

FEXCore/include/FEXCore/IR/IR.h

@@ -3,31 +3,12 @@
 
 #include <FEXCore/Utils/EnumOperators.h>
 
+#include <compare>
 #include <cstdint>
 #include <cstring>
 
 namespace FEXCore::IR {
 
-enum class SyscallFlags : uint8_t {
-  DEFAULT = 0,
-  // Syscalldoesn't care about CPUState being serialized up to the syscall instruction.
-  // Means dead code elimination can optimize through a syscall operation.
-  OPTIMIZETHROUGH = 1 << 0,
-  // Syscall only reads the passed in arguments. Doesn't read CPUState.
-  NOSYNCSTATEONENTRY = 1 << 1,
-  // Syscall doesn't return. Code generation after syscall return can be removed.
-  NORETURN = 1 << 2,
-  // Syscall doesn't have any side-effects, so if the result isn't used then it can be removed.
-  NOSIDEEFFECTS = 1 << 3,
-  // Syscall doesn't return a result.
-  // Means the resulting register shouldn't be written (Usually RAX).
-  // Usually used with !NOSYNCSTATEONENTRY, so the syscall can modify CPU state entirely.
-  // Then on return FEXCore picks up the new state.
-  NORETURNEDRESULT = 1 << 4,
-};
-
-FEX_DEF_NUM_OPS(SyscallFlags)
-
 // This enum of named vector constants are linked to an array in CPUBackend.cpp.
 // This is used with the IROp `LoadNamedVectorConstant` to load a vector constant
 // that would otherwise be costly to materialize.
@@ -96,15 +77,7 @@ enum IndexNamedVectorConstant : uint8_t {
 
 struct SHA256Sum final {
   uint8_t data[32];
-  [[nodiscard]]
-  bool operator<(const SHA256Sum& rhs) const {
-    return memcmp(data, rhs.data, sizeof(data)) < 0;
-  }
-
-  [[nodiscard]]
-  bool operator==(const SHA256Sum& rhs) const {
-    return memcmp(data, rhs.data, sizeof(data)) == 0;
-  }
+  [[nodiscard]] auto operator<=>(const SHA256Sum&) const noexcept = default;
 };
 
 typedef void ThunkedFunction(void* ArgsRv);

FEXCore/include/FEXCore/Utils/AllocatorHooks.h

@@ -82,12 +82,15 @@ inline bool VirtualProtect(void* Ptr, size_t Size, ProtectOptions options) {
   return ::VirtualProtect(Ptr, Size, prot, nullptr) == 0;
 }
 
+inline void VirtualName(const char*, void*, size_t) {}
+
 #else
 using MMAP_Hook = void* (*)(void*, size_t, int, int, int, off_t);
 using MUNMAP_Hook = int (*)(void*, size_t);
 
 FEX_DEFAULT_VISIBILITY extern MMAP_Hook mmap;
 FEX_DEFAULT_VISIBILITY extern MUNMAP_Hook munmap;
+FEX_DEFAULT_VISIBILITY extern void VirtualName(const char* Name, void* Ptr, size_t Size);
 
 // All commit parameters are ignored here, they are unnecessary as Linux supports overcommit
 

FEXCore/include/FEXCore/Utils/ArchHelpers/Arm64.h

@@ -12,8 +12,6 @@ struct InternalThreadState;
 
 namespace FEXCore::ArchHelpers::Arm64 {
 enum class UnalignedHandlerType {
-  ///< Don't backpatch code, instead handle inside SIGBUS handler.
-  Paranoid,
   ///< Backpatch unaligned access to half-barrier based atomic.
   HalfBarrier,
   ///< Backpatch unaligned access to non-atomic.
@@ -26,7 +24,7 @@ enum class UnalignedHandlerType {
  * This is an OS agnostic handler where the frontend must provide FEXCore with the information necessary to know if this is safe.
  * This does not check if the PC is within a JIT code buffer, the frontend must provide that safety with `CPUBackend::IsAddressInCodeBuffer`.
  *
- * @param ParanoidTSO If the unaligned fault needs to handled directly or can be backpatched.
+ * @param HandleType Type of TSO handling to use.
  * @param ProgramCounter The location in memory for the instruction that did the access
  * @param GPRs The array of GPRs from the signal context. This will be modified and the host context needs to be updated on signal return.
  *
@@ -34,6 +32,6 @@ enum class UnalignedHandlerType {
  * by. FEXCore will return a positive or negative offset depending on internal handling.
  */
 [[nodiscard]]
-FEX_DEFAULT_VISIBILITY std::optional<int32_t>
-HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandlerType HandleType, uintptr_t ProgramCounter, uint64_t* GPRs);
+FEX_DEFAULT_VISIBILITY std::optional<int32_t> HandleUnalignedAccess(
+  FEXCore::Core::InternalThreadState* Thread, UnalignedHandlerType HandleType, uintptr_t ProgramCounter, uint64_t* GPRs, bool IsJIT = true);
 } // namespace FEXCore::ArchHelpers::Arm64

FEXCore/include/FEXCore/Utils/EnumUtils.h

@@ -54,6 +54,13 @@ namespace FEXCore {
     return static_cast<T>(key) == 0;                                 \
   }
 
+// Macro that defines a fmt formatter for a reasonable case where an enum
+// is formatted as a purely integral type based on its underlying type.
+#define FEX_DEFINE_ENUM_FMT_PASSTHROUGH(type) \
+  constexpr auto format_as(type t) {          \
+    return FEXCore::ToUnderlying(t);          \
+  }
+
 // Equivalent to C++23's std::to_underlying.
 template<typename Enum>
 [[nodiscard]]

FEXCore/include/FEXCore/Utils/LongJump.h

@@ -4,8 +4,10 @@
 
 #include <cstdint>
 
-// It's longjump without glibc fortification checks.
-namespace FEXCore::LongJump {
+// Reimplementation of longjmp without glibc fortification checks.
+// This is useful when false positives need to be avoided or when using
+// a libc implementation that does not implement std::longjmp.
+namespace FEXCore::UncheckedLongJump {
 // JumpBuf definition needs to be public because the frontend needs to understand it.
 #if defined(_M_ARM_64)
 struct JumpBuf {
@@ -32,5 +34,6 @@ struct JumpBuf {
 #endif
 
 [[nodiscard]] FEX_DEFAULT_VISIBILITY uint64_t SetJump(JumpBuf& Buffer);
-[[noreturn]] FEX_DEFAULT_VISIBILITY void LongJump(JumpBuf& Buffer, uint64_t Value);
-} // namespace FEXCore::LongJump
+[[noreturn]] FEX_DEFAULT_VISIBILITY void LongJump(const JumpBuf& Buffer, uint64_t Value);
+FEX_DEFAULT_VISIBILITY void ManuallyLoadJumpBuf(const JumpBuf& Buffer, uint64_t Value, uint64_t* GPRs, __uint128_t* FPRs, uint64_t* PC);
+} // namespace FEXCore::UncheckedLongJump

FEXCore/include/FEXCore/Utils/PrctlUtils.h

@@ -0,0 +1,54 @@
+// SPDX-License-Identifier: MIT
+#pragma once
+
+#ifndef _WIN32
+#include <linux/prctl.h>
+#include <sys/mman.h>
+#include <sys/user.h>
+#include <sys/prctl.h>
+
+#ifndef PR_SET_VMA
+#define PR_SET_VMA 0x53564d41
+#endif
+
+#ifndef PR_SET_VMA_ANON_NAME
+#define PR_SET_VMA_ANON_NAME 0
+#endif
+
+#ifndef PR_GET_MEM_MODEL
+#define PR_GET_MEM_MODEL 0x6d4d444c
+#endif
+#ifndef PR_SET_MEM_MODEL
+#define PR_SET_MEM_MODEL 0x4d4d444c
+#endif
+#ifndef PR_SET_MEM_MODEL_DEFAULT
+#define PR_SET_MEM_MODEL_DEFAULT 0
+#endif
+#ifndef PR_SET_MEM_MODEL_TSO
+#define PR_SET_MEM_MODEL_TSO 1
+#endif
+
+#ifndef PR_GET_COMPAT_INPUT
+#define PR_GET_COMPAT_INPUT 0x63494e50
+#endif
+#ifndef PR_SET_COMPAT_INPUT
+#define PR_SET_COMPAT_INPUT 0x43494e50
+#endif
+#ifndef PR_SET_COMPAT_INPUT_DISABLE
+#define PR_SET_COMPAT_INPUT_DISABLE 0
+#endif
+#ifndef PR_SET_COMPAT_INPUT_ENABLE
+#define PR_SET_COMPAT_INPUT_ENABLE 1
+#endif
+
+#ifndef PR_GET_SHADOW_STACK_STATUS
+#define PR_GET_SHADOW_STACK_STATUS 74
+#endif
+#ifndef PR_LOCK_SHADOW_STACK_STATUS
+#define PR_LOCK_SHADOW_STACK_STATUS 76
+#endif
+#ifndef PR_SHADOW_STACK_ENABLE
+#define PR_SHADOW_STACK_ENABLE (1ULL << 0)
+#endif
+
+#endif // ifndef _WIN32