Docs: Update for release FEX-2510

SyscallHandler: Shrink definition struct from 24 to 16 bytes

.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

CMakeLists.txt

@@ -319,7 +319,7 @@ 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)
 

CodeEmitter/CodeEmitter/ALUOps.inl

@@ -36,33 +36,24 @@ public:
     DataProcessing_PCRel_Imm(Op, rd, Imm);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, const BackwardLabel* Label) {
+  void 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]] {
-      constexpr uint32_t Op = 0b0001'0000 << 24;
-      DataProcessing_PCRel_Imm(Op, rd, Imm);
-      return BranchEncodeSucceeded::Success;
-    }
-
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    constexpr uint32_t Op = 0b0001'0000 << 24;
+    DataProcessing_PCRel_Imm(Op, rd, Imm);
   }
-  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, ForwardLabel* Label) {
+  void adr(ARMEmitter::Register rd, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::ADR});
     constexpr uint32_t Op = 0b0001'0000 << 24;
     DataProcessing_PCRel_Imm(Op, rd, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  void adr(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return adr(rd, &Label->Backward);
+      adr(rd, &Label->Backward);
     } else {
-      return adr(rd, &Label->Forward);
+      adr(rd, &Label->Forward);
     }
   }
 
@@ -71,42 +62,32 @@ public:
     DataProcessing_PCRel_Imm(Op, rd, Imm);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, const BackwardLabel* Label) {
+  void 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]] {
-      constexpr uint32_t Op = 0b1001'0000 << 24;
-      DataProcessing_PCRel_Imm(Op, rd, Imm);
-      return BranchEncodeSucceeded::Success;
-    }
-
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    constexpr uint32_t Op = 0b1001'0000 << 24;
+    DataProcessing_PCRel_Imm(Op, rd, Imm);
   }
-
-  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, ForwardLabel* Label) {
+  void adrp(ARMEmitter::Register rd, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::ADRP});
     constexpr uint32_t Op = 0b1001'0000 << 24;
     DataProcessing_PCRel_Imm(Op, rd, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  void adrp(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return adrp(rd, &Label->Backward);
+      adrp(rd, &Label->Backward);
     } else {
-      return adrp(rd, &Label->Forward);
+      adrp(rd, &Label->Forward);
     }
   }
 
-  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, const BackwardLabel* Label) {
+  void LongAddressGen(ARMEmitter::Register rd, const BackwardLabel* Label) {
     int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>());
     if (IsADRRange(Imm)) {
       // If the range is in ADR range then we can just use ADR.
-      return adr(rd, Label);
+      adr(rd, Label);
     } else if (IsADRPRange(Imm)) {
       int64_t ADRPImm = (reinterpret_cast<int64_t>(Label->Location) & ~0xFFFLL) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
 
@@ -121,28 +102,23 @@ public:
         // Now even an add
         add(ARMEmitter::Size::i64Bit, rd, rd, AlignedOffset);
       }
-
-      return BranchEncodeSucceeded::Success;
+    } else {
+      LOGMAN_MSG_A_FMT("Unscaled offset too large");
+      FEX_UNREACHABLE;
     }
-
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
   }
-  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, ForwardLabel* Label) {
+  void 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.
     dc32(rd.Idx());
     nop();
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  void LongAddressGen(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return LongAddressGen(rd, &Label->Backward);
+      LongAddressGen(rd, &Label->Backward);
     } else {
-      return LongAddressGen(rd, &Label->Forward);
+      LongAddressGen(rd, &Label->Forward);
     }
   }
 
@@ -886,6 +862,12 @@ public:
   }
 
 private:
+  static constexpr Condition InvertCondition(Condition cond) {
+    // These behave as always, so it makes no sense to allow inverting these.
+    LOGMAN_THROW_A_FMT(cond != Condition::CC_AL && cond != Condition::CC_NV, "Cannot invert CC_AL or CC_NV");
+    return static_cast<Condition>(FEXCore::ToUnderlying(cond) ^ 1);
+  }
+
   void and_(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t n, uint32_t immr, uint32_t imms) {
     constexpr uint32_t Op = 0b001'0010'00 << 22;
     DataProcessing_Logical_Imm(Op, s, rd, rn, n, immr, imms);

CodeEmitter/CodeEmitter/BranchOps.inl

@@ -20,31 +20,23 @@ public:
     constexpr uint32_t Op = 0b0101'010 << 25;
     Branch_Conditional(Op, 0, 0, Cond, Imm);
   }
-  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
+  void 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]] {
-      constexpr uint32_t Op = 0b0101'010 << 25;
-      Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
-      return BranchEncodeSucceeded::Success;
-    }
-
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+    constexpr uint32_t Op = 0b0101'010 << 25;
+    Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
   }
-  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, ForwardLabel* Label) {
+  void b(ARMEmitter::Condition Cond, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
     constexpr uint32_t Op = 0b0101'010 << 25;
     Branch_Conditional(Op, 0, 0, Cond, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
+  void b(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return b(Cond, &Label->Backward);
+      b(Cond, &Label->Backward);
     } else {
-      return b(Cond, &Label->Forward);
+      b(Cond, &Label->Forward);
     }
   }
 
@@ -53,32 +45,24 @@ public:
     constexpr uint32_t Op = 0b0101'010 << 25;
     Branch_Conditional(Op, 0, 1, Cond, Imm);
   }
-  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
+  void 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]] {
-      constexpr uint32_t Op = 0b0101'010 << 25;
-      Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
-      return BranchEncodeSucceeded::Success;
-    }
-
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+    constexpr uint32_t Op = 0b0101'010 << 25;
+    Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, ForwardLabel* Label) {
+  void bc(ARMEmitter::Condition Cond, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
     constexpr uint32_t Op = 0b0101'010 << 25;
     Branch_Conditional(Op, 0, 1, Cond, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
+  void bc(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return bc(Cond, &Label->Backward);
+      bc(Cond, &Label->Backward);
     } else {
-      return bc(Cond, &Label->Forward);
+      bc(Cond, &Label->Forward);
     }
   }
 
@@ -114,32 +98,25 @@ public:
 
     UnconditionalBranch(Op, Imm);
   }
-  [[nodiscard]] BranchEncodeSucceeded b(const BackwardLabel* Label) {
+  void 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]] {
-      constexpr uint32_t Op = 0b0001'01 << 26;
-      UnconditionalBranch(Op, Imm >> 2);
-      return BranchEncodeSucceeded::Success;
-    }
+    LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+    constexpr uint32_t Op = 0b0001'01 << 26;
 
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    UnconditionalBranch(Op, Imm >> 2);
   }
-  [[nodiscard]] BranchEncodeSucceeded b(ForwardLabel* Label) {
+  void b(ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::B});
     constexpr uint32_t Op = 0b0001'01 << 26;
 
     UnconditionalBranch(Op, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded b(BiDirectionalLabel* Label) {
+  void b(BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return b(&Label->Backward);
+      b(&Label->Backward);
     } else {
-      return b(&Label->Forward);
+      b(&Label->Forward);
     }
   }
 
@@ -149,33 +126,25 @@ public:
     UnconditionalBranch(Op, Imm);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded bl(const BackwardLabel* Label) {
+  void 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]] {
-      constexpr uint32_t Op = 0b1001'01 << 26;
-      UnconditionalBranch(Op, Imm >> 2);
+    LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+    constexpr uint32_t Op = 0b1001'01 << 26;
 
-      return BranchEncodeSucceeded::Success;
-    }
-
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    UnconditionalBranch(Op, Imm >> 2);
   }
-  [[nodiscard]] BranchEncodeSucceeded bl(ForwardLabel* Label) {
+  void bl(ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::B});
     constexpr uint32_t Op = 0b1001'01 << 26;
 
     UnconditionalBranch(Op, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded bl(BiDirectionalLabel* Label) {
+  void bl(BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return bl(&Label->Backward);
+      bl(&Label->Backward);
     } else {
-      return bl(&Label->Forward);
+      bl(&Label->Forward);
     }
   }
 
@@ -186,35 +155,28 @@ public:
     CompareAndBranch(Op, s, rt, Imm);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
+  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
 
-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
-      constexpr uint32_t Op = 0b0011'0100 << 24;
-      CompareAndBranch(Op, s, rt, Imm >> 2);
-      return BranchEncodeSucceeded::Success;
-    }
+    constexpr uint32_t Op = 0b0011'0100 << 24;
 
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    CompareAndBranch(Op, s, rt, Imm >> 2);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
+  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
 
     constexpr uint32_t Op = 0b0011'0100 << 24;
 
     CompareAndBranch(Op, s, rt, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
+  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return cbz(s, rt, &Label->Backward);
+      cbz(s, rt, &Label->Backward);
     } else {
-      return cbz(s, rt, &Label->Forward);
+      cbz(s, rt, &Label->Forward);
     }
   }
 
@@ -224,35 +186,28 @@ public:
     CompareAndBranch(Op, s, rt, Imm);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
+  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
 
-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
-      constexpr uint32_t Op = 0b0011'0101 << 24;
-      CompareAndBranch(Op, s, rt, Imm >> 2);
-      return BranchEncodeSucceeded::Success;
-    }
+    constexpr uint32_t Op = 0b0011'0101 << 24;
 
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    CompareAndBranch(Op, s, rt, Imm >> 2);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
+  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
 
     constexpr uint32_t Op = 0b0011'0101 << 24;
 
     CompareAndBranch(Op, s, rt, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
+  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return cbnz(s, rt, &Label->Backward);
+      cbnz(s, rt, &Label->Backward);
     } else {
-      return cbnz(s, rt, &Label->Forward);
+      cbnz(s, rt, &Label->Forward);
     }
   }
 
@@ -262,35 +217,28 @@ public:
 
     TestAndBranch(Op, rt, Bit, Imm);
   }
-  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
+  void tbz(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]] {
-      constexpr uint32_t Op = 0b0011'0110 << 24;
-      TestAndBranch(Op, rt, Bit, Imm >> 2);
-      return BranchEncodeSucceeded::Success;
-    }
+    constexpr uint32_t Op = 0b0011'0110 << 24;
 
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    TestAndBranch(Op, rt, Bit, Imm >> 2);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
+  void tbz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::TEST_BRANCH});
 
     constexpr uint32_t Op = 0b0011'0110 << 24;
 
     TestAndBranch(Op, rt, Bit, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
+  void tbz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return tbz(rt, Bit, &Label->Backward);
+      tbz(rt, Bit, &Label->Backward);
     } else {
-      return tbz(rt, Bit, &Label->Forward);
+      tbz(rt, Bit, &Label->Forward);
     }
   }
 
@@ -299,35 +247,27 @@ public:
 
     TestAndBranch(Op, rt, Bit, Imm);
   }
-  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
+  void 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]] {
-      constexpr uint32_t Op = 0b0011'0111 << 24;
-      TestAndBranch(Op, rt, Bit, Imm >> 2);
-      return BranchEncodeSucceeded::Success;
-    }
+    constexpr uint32_t Op = 0b0011'0111 << 24;
 
-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    TestAndBranch(Op, rt, Bit, Imm >> 2);
   }
 
-  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
+  void tbnz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::TEST_BRANCH});
     constexpr uint32_t Op = 0b0011'0111 << 24;
 
     TestAndBranch(Op, rt, Bit, 0);
-
-    // Forward label doesn't know if it can encode until Bind.
-    return BranchEncodeSucceeded::Success;
   }
 
-  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
+  void tbnz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      return tbnz(rt, Bit, &Label->Backward);
+      tbnz(rt, Bit, &Label->Backward);
     } else {
-      return tbnz(rt, Bit, &Label->Forward);
+      tbnz(rt, Bit, &Label->Forward);
     }
   }
 

CodeEmitter/CodeEmitter/Emitter.h

@@ -586,11 +586,6 @@ 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>;
 
-enum class BranchEncodeSucceeded {
-  Success,
-  Failure,
-};
-
 // Whether or not a given set of vector registers are sequential
 // in increasing order as far as the register file is concerned (modulo its size)
 //
@@ -643,25 +638,19 @@ public:
 
   // Bind a backward label to an address.
   // Address that is bound is the current emitter location.
-  [[nodiscard]] bool Bind(BackwardLabel* Label) {
+  void Bind(BackwardLabel* Label) {
     LOGMAN_THROW_A_FMT(Label->Location == nullptr, "Trying to bind a label twice");
     Label->Location = GetCursorAddress<uint8_t*>();
-
-    // Always binds because it is only storing a location.
-    return true;
   }
 
-  [[nodiscard]] bool Bind(const ForwardLabel::Reference* Label) {
+  void Bind(const ForwardLabel::Reference* Label) {
     uint8_t* CurrentAddress = GetCursorAddress<uint8_t*>();
     // Patch up the instructions
     switch (Label->Type) {
     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]] {
-        // Can't bind.
-        return false;
-      }
+      LOGMAN_THROW_A_FMT(IsADRRange(Imm), "Unscaled offset too large");
       uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
       uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
       uint32_t Inst = *Instruction & ~InstMask;
@@ -673,12 +662,7 @@ public:
     case ForwardLabel::InstType::ADRP: {
       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]] {
-        // Can't bind.
-        return false;
-      }
-
+      LOGMAN_THROW_A_FMT(IsADRPRange(Imm) && IsADRPAligned(Imm), "Unscaled offset too large");
       Imm >>= 12;
       uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
       uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
@@ -688,13 +672,11 @@ public:
       *Instruction = Inst;
       break;
     }
+
     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]] {
-        // Can't bind.
-        return false;
-      }
+      LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
       Imm >>= 2;
       uint32_t InstMask = 0x3FF'FFFF;
       uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -704,13 +686,11 @@ public:
 
       break;
     }
+
     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]] {
-        // Can't bind.
-        return false;
-      }
+      LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
       Imm >>= 2;
       uint32_t InstMask = 0x3FFF;
       uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -724,10 +704,7 @@ 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]] {
-        // Can't bind.
-        return false;
-      }
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
       Imm >>= 2;
       uint32_t InstMask = 0x7'FFFF;
       uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -776,41 +753,27 @@ public:
     }
     default: LOGMAN_MSG_A_FMT("Unexpected inst type in label fixup");
     }
-
-    return true;
   }
 
   // Bind a forward label to a location.
   // This walks all the instructions in the label's vector.
   // Then backpatching all instructions that have used the label.
-  [[nodiscard]] bool Bind(ForwardLabel* Label) {
-    bool Bound = true;
+  void Bind(ForwardLabel* Label) {
     if (Label->FirstInst.Location) {
-      Bound &= Bind(&Label->FirstInst);
+      Bind(&Label->FirstInst);
     }
     for (auto& Inst : Label->Insts) {
-      Bound &= Bind(&Inst);
+      Bind(&Inst);
     }
-
-    return Bound;
   }
 
   // Bind a bidirectional location to a location.
   // Binds both forwards and backwards depending on how the label was used.
-  [[nodiscard]] bool Bind(BiDirectionalLabel* Label) {
-    bool Bound = true;
+  void Bind(BiDirectionalLabel* Label) {
     if (!Label->Backward.Location) {
-      Bound &= Bind(&Label->Backward);
+      Bind(&Label->Backward);
     }
-    Bound &= Bind(&Label->Forward);
-
-    return Bound;
-  }
-
-  static constexpr Condition InvertCondition(Condition cond) {
-    // These behave as always, so it makes no sense to allow inverting these.
-    LOGMAN_THROW_A_FMT(cond != Condition::CC_AL && cond != Condition::CC_NV, "Cannot invert CC_AL or CC_NV");
-    return static_cast<Condition>(FEXCore::ToUnderlying(cond) ^ 1);
+    Bind(&Label->Forward);
   }
 
 #include <CodeEmitter/VixlUtils.inl>

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/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

@@ -66,7 +66,6 @@ set (SRCS
   Interface/IR/Passes/RedundantFlagCalculationElimination.cpp
   Interface/IR/Passes/RegisterAllocationPass.cpp
   Interface/IR/Passes/x87StackOptimizationPass.cpp
-  Utils/LongJump.cpp
   Utils/Telemetry.cpp
   Utils/Threads.cpp
   Utils/Profiler.cpp

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 +501,53 @@ 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);
+  }
+
+  bool IsSignalingNaN() const {
+    return (Exponent == 0x7FFF) && (Significand & 0x8000000000000000ULL) && !(Significand & 0x4000000000000000ULL) && // Bit 62 clear (signaling)
+           (Significand & 0x3FFFFFFFFFFFFFFFULL);
+  }
+
+  bool IsQuietNaN() const {
+    return (Exponent == 0x7FFF) && (Significand & 0x8000000000000000ULL) && (Significand & 0x4000000000000000ULL); // Bit 62 set (quiet)
+  }
+
+  // Helper to detect if this is any NaN
+  bool IsNaN() const {
+    return IsSignalingNaN() || IsQuietNaN();
+  }
+
+  // X87 value to F64 while preserving signaling nan property
+  double ToF64_PreserveNan(softfloat_state* state) const {
+    if (IsSignalingNaN()) {
+      // we keep it as a signaling nan in ieee754 in 64bits
+      uint64_t sign_bit = Sign ? 0x8000000000000000ULL : 0;
+      uint64_t exp_bits = 0x7FF0000000000000ULL;
+      uint64_t x87_frac = Significand & 0x3FFFFFFFFFFFFFFFULL;
+      uint64_t ieee_frac = (x87_frac >> 11) & 0x0007FFFFFFFFFFFFULL;
+
+      if (ieee_frac == 0) {
+        ieee_frac = 1;
+      }
+      ieee_frac &= ~0x0008000000000000ULL;
+
+      uint64_t result_bits = sign_bit | exp_bits | ieee_frac;
+      return std::bit_cast<double>(result_bits);
+    } else if (IsQuietNaN()) {
+      const float64_t Result = extF80_to_f64(state, *this);
+      uint64_t result_bits = std::bit_cast<uint64_t>(Result);
+      result_bits |= 0x0008000000000000ULL;
+      return std::bit_cast<double>(result_bits);
+    } else {
+      const float64_t Result = extF80_to_f64(state, *this);
+      return std::bit_cast<double>(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 +559,7 @@ 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 +618,51 @@ 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));
+  }
+
+  // Create X80SoftFloat from double while preserving NaN signaling properties
+  static X80SoftFloat FromF64_PreserveNaN(softfloat_state* state, double value) {
+    uint64_t bits = std::bit_cast<uint64_t>(value);
+
+    // Check if it's a nan
+    if ((bits & 0x7FF0000000000000ULL) == 0x7FF0000000000000ULL && (bits & 0x000FFFFFFFFFFFFFULL) != 0) {
+
+      X80SoftFloat result;
+      result.Sign = (bits >> 63) & 1;
+      result.Exponent = 0x7FFF;
+
+      bool is_signaling = !(bits & 0x0008000000000000ULL);
+      uint64_t ieee_payload = bits & 0x0007FFFFFFFFFFFFULL;
+
+      // set bit 63 required for x87
+      result.Significand = 0x8000000000000000ULL;
+
+      if (is_signaling) { // clear bit 62 for signaling nan
+        result.Significand &= ~0x4000000000000000ULL;
+      } else { // clear bit 62 for quiet nan
+        result.Significand |= 0x4000000000000000ULL;
+      }
+
+      // ieee754 51-bit payload -> x87 62-bit payload
+      result.Significand |= (ieee_payload << 11) & 0x3FFFFFFFFFFFFFFFULL;
+
+      return result;
+    }
+
+    // For non-NaN values, use standard conversion
+    return X80SoftFloat(state, value);
   }
 
   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/Common/VectorRegType.h

@@ -4,8 +4,6 @@
 #ifdef _M_X86_64
 #include <xmmintrin.h>
 #include <immintrin.h>
-#else
-#include <cstdint>
 #endif
 
 namespace FEXCore {

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

@@ -59,7 +59,9 @@
           "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 +84,8 @@
           "\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": {
@@ -327,13 +330,6 @@
           "Enables FEX's low-overhead sampling profile statistics.",
           "Requires a supported version of Mangohud to see the results"
         ]
-      },
-      "TraceProfiler": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "Enables FEX's trace profiler. Using gpuvis or tracy"
-        ]
       }
     },
     "Hacks": {
@@ -403,6 +399,14 @@
           "Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
         ]
       },
+      "X87StrictReducedPrecision": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enables stricter X87 floating point behavior when X87ReducedPrecision is enabled.",
+          "Adds additional checks and implementations like NaN propagation for better compatibility."
+        ]
+      },
       "ABILocalFlags": {
         "Type": "bool",
         "Default": "false",

FEXCore/Source/Interface/Context/Context.h

@@ -207,6 +207,7 @@ public:
     FEX_CONFIG_OPT(GDBSymbols, GDBSYMBOLS);
     FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
     FEX_CONFIG_OPT(x87ReducedPrecision, X87REDUCEDPRECISION);
+    FEX_CONFIG_OPT(x87StrictReducedPrecision, X87STRICTREDUCEDPRECISION);
     FEX_CONFIG_OPT(DisableTelemetry, DISABLETELEMETRY);
     FEX_CONFIG_OPT(DisableVixlIndirectCalls, DISABLE_VIXL_INDIRECT_RUNTIME_CALLS);
     FEX_CONFIG_OPT(SmallTSCScale, SMALLTSCSCALE);
@@ -322,6 +323,13 @@ protected:
     }
   }
 
+  void UpdateX87PrecisionConfig() {
+    // If strict reduced precision is enabled, automatically enable reduced precision
+    if (Config.x87StrictReducedPrecision() && !Config.x87ReducedPrecision()) {
+      FEXCore::Config::Set(FEXCore::Config::CONFIG_X87REDUCEDPRECISION, "1");
+    }
+  }
+
 private:
   /**
    * @brief Initializes the JIT compilers for the thread

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

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 +360,10 @@ namespace CPU {
       LogMan::Msg::EFmt("Failed to mprotect last page of code buffer.");
     }
 
+#ifndef _WIN32
+    prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, Ptr, Size, "FEXMemJIT");
+#endif
+
     LookupCache = fextl::make_unique<GuestToHostMap>();
   }
 

FEXCore/Source/Interface/Core/CPUID.cpp

@@ -909,38 +909,38 @@ 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

FEXCore/Source/Interface/Core/Core.cpp

@@ -57,20 +57,13 @@ $end_info$
 #include <algorithm>
 #include <array>
 #include <atomic>
-#include <chrono>
-#include <condition_variable>
 #include <fcntl.h>
 #include <functional>
 #include <mutex>
-#include <queue>
 #include <shared_mutex>
-#include <signal.h>
 #include <stdio.h>
-#include <string_view>
 #include <sys/stat.h>
-#include <type_traits>
 #include <unistd.h>
-#include <unordered_map>
 #include <utility>
 #include <xxhash.h>
 
@@ -100,6 +93,8 @@ ContextImpl::ContextImpl(const FEXCore::HostFeatures& Features)
 
   // Track atomic TSO emulation configuration.
   UpdateAtomicTSOEmulationConfig();
+  // Ensure X87 precision constraints are respected.
+  UpdateX87PrecisionConfig();
 }
 
 struct GetFrameBlockInfoResult {
@@ -952,10 +947,10 @@ 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());
     },

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

@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: MIT
 
-#include "Common/VectorRegType.h"
+#include "Common/SoftFloat.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
@@ -26,7 +26,9 @@
 #endif
 
 #include <array>
+#include <atomic>
 #include <bit>
+#include <condition_variable>
 #include <csignal>
 #include <cstring>
 
@@ -91,7 +93,7 @@ void Dispatcher::EmitDispatcher() {
 
   FillStaticRegs();
   ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));
-  (void)cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
+  cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
 
   ARMEmitter::BiDirectionalLabel LoopTop {};
 
@@ -140,7 +142,7 @@ void Dispatcher::EmitDispatcher() {
   // We want to ensure that we are 16 byte aligned at the top of this loop
   Align16B();
 
-  (void)Bind(&LoopTop);
+  Bind(&LoopTop);
   AbsoluteLoopTopAddress = GetCursorAddress<uint64_t>();
 
   // Load in our RIP
@@ -167,11 +169,11 @@ void Dispatcher::EmitDispatcher() {
   ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionEC));
   br(TMP2);
 
-  (void)!Bind(&l_NotECCode);
+  Bind(&l_NotECCode);
 #endif
 
   ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-  (void)cbnz(ARMEmitter::Size::i32Bit, TMP1, &CompileSingleStep);
+  cbnz(ARMEmitter::Size::i32Bit, TMP1, &CompileSingleStep);
 
   // This is the block cache lookup routine
   // It matches what is going on it LookupCache.h::FindBlock
@@ -196,7 +198,7 @@ void Dispatcher::EmitDispatcher() {
     ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 3);
 
     // If page pointer is zero then we have no block
-    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &NoBlock);
+    cbz(ARMEmitter::Size::i64Bit, TMP1, &NoBlock);
 
     // Steal the page offset
     and_(ARMEmitter::Size::i64Bit, TMP2, TMP4, 0x0FFF);
@@ -211,11 +213,10 @@ void Dispatcher::EmitDispatcher() {
 
     // If the guest address doesn't match, Compile the block.
     sub(TMP2, TMP2, RipReg);
-    (void)cbnz(ARMEmitter::Size::i64Bit, TMP2, &NoBlock);
-
+    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);
+    cbz(ARMEmitter::Size::i64Bit, TMP4, &NoBlock);
 
     // If we've made it here then we have a real compiled block
     {
@@ -303,7 +304,7 @@ void Dispatcher::EmitDispatcher() {
 
   // Need to create the block
   {
-    (void)Bind(&NoBlock);
+    Bind(&NoBlock);
 
     EmitSignalGuardedRegion([&]() {
       SpillStaticRegs(TMP1);
@@ -337,7 +338,7 @@ void Dispatcher::EmitDispatcher() {
   }
 
   {
-    (void)Bind(&CompileSingleStep);
+    Bind(&CompileSingleStep);
 
     EmitSignalGuardedRegion([&]() {
       SpillStaticRegs(TMP1);
@@ -499,7 +500,7 @@ void Dispatcher::EmitDispatcher() {
     stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
 
     // Now go back to the regular dispatcher loop
-    (void)b(&LoopTop);
+    b(&LoopTop);
   }
 
   auto EmitLongALUOpHandler = [&](auto R, auto Offset) {
@@ -568,15 +569,14 @@ void Dispatcher::EmitDispatcher() {
     }
   }
 
-  (void)Bind(&l_CTX);
+  Bind(&l_CTX);
   dc64(reinterpret_cast<uintptr_t>(CTX));
-  (void)Bind(&l_Sleep);
+  Bind(&l_Sleep);
   dc64(reinterpret_cast<uint64_t>(SleepThread));
-  (void)Bind(&l_CompileBlock);
+  Bind(&l_CompileBlock);
   FEXCore::Utils::MemberFunctionToPointerCast PMFCompileBlock(&FEXCore::Context::ContextImpl::CompileBlock);
   dc64(PMFCompileBlock.GetConvertedPointer());
-  (void)Bind(&l_CompileSingleStep);
-
+  Bind(&l_CompileSingleStep);
   FEXCore::Utils::MemberFunctionToPointerCast PMFCompileSingleStep(&FEXCore::Context::ContextImpl::CompileSingleStep);
   dc64(PMFCompileSingleStep.GetConvertedPointer());
 

FEXCore/Source/Interface/Core/Interpreter/Fallbacks/F80Fallbacks.h

@@ -2,11 +2,13 @@
 #pragma once
 #include "Common/SoftFloat.h"
 
+#include "Interface/Context/Context.h"
 #include "Interface/Core/Interpreter/Fallbacks/FallbackOpHandler.h"
 #include "Interface/IR/IR.h"
 
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/SHMStats.h>
+#include <FEXCore/Config/Config.h>
 
 namespace FEXCore::CPU {
 FEXCORE_PRESERVE_ALL_ATTR static softfloat_state SoftFloatStateFromFCW(uint16_t FCW, bool Force80BitPrecision = false) {
@@ -77,6 +79,12 @@ struct OpHandlers<IR::OP_F80CVTTO> {
   FEXCORE_PRESERVE_ALL_ATTR static VectorRegType handle8(uint16_t FCW, double src, FEXCore::Core::CpuStateFrame* Frame) {
     FEXCORE_PROFILE_INSTANT_INCREMENT(Frame->Thread, AccumulatedFloatFallbackCount, 1);
     ScopedSoftFloatState State {FCW, Frame};
+    auto Context = static_cast<Context::ContextImpl*>(Frame->Thread->CTX);
+    auto ReducedPrecisionMode = Context->Config.x87ReducedPrecision;
+    auto StrictReducedPrecisionMode = Context->Config.x87StrictReducedPrecision;
+    if (!ReducedPrecisionMode || StrictReducedPrecisionMode) {
+      return X80SoftFloat::FromF64_PreserveNaN(&State.State, src);
+    }
     return X80SoftFloat(&State.State, src);
   }
 };
@@ -115,6 +123,12 @@ struct OpHandlers<IR::OP_F80CVT> {
   FEXCORE_PRESERVE_ALL_ATTR static double handle8(uint16_t FCW, VectorRegType src, FEXCore::Core::CpuStateFrame* Frame) {
     FEXCORE_PROFILE_INSTANT_INCREMENT(Frame->Thread, AccumulatedFloatFallbackCount, 1);
     ScopedSoftFloatState State {FCW, Frame};
+    auto Context = static_cast<Context::ContextImpl*>(Frame->Thread->CTX);
+    auto ReducedPrecisionMode = Context->Config.x87ReducedPrecision;
+    auto StrictReducedPrecisionMode = Context->Config.x87StrictReducedPrecision;
+    if (!ReducedPrecisionMode || StrictReducedPrecisionMode) {
+      return X80SoftFloat(src).ToF64_PreserveNan(&State.State);
+    }
     return X80SoftFloat(src).ToF64(&State.State);
   }
 };

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

@@ -372,7 +372,7 @@ 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));
@@ -588,7 +588,7 @@ DEF_OP(ShiftFlags) {
   and_(ARMEmitter::Size::i32Bit, TMP1, Src2, OpSize == IR::OpSize::i64Bit ? 0x3f : 0x1f);
 
   ARMEmitter::ForwardLabel Done;
-  (void)cbz(EmitSize, TMP1, &Done);
+  cbz(EmitSize, TMP1, &Done);
   {
     // PF/SF/ZF/OF
     if (OpSize >= IR::OpSize::i32Bit) {
@@ -652,7 +652,7 @@ DEF_OP(ShiftFlags) {
       msr(ARMEmitter::SystemRegister::NZCV, TMP2);
     }
   }
-  (void)Bind(&Done);
+  Bind(&Done);
 
   // TODO: Make RA less dumb so this can't happen (e.g. with late-kill).
   if (PFOutput != PFTemp) {
@@ -669,7 +669,7 @@ DEF_OP(RotateFlags) {
 
   // If shift=0, flags are unaffected. Wrap the whole implementation in a cbz.
   ARMEmitter::ForwardLabel Done;
-  (void)cbz(EmitSize, Shift, &Done);
+  cbz(EmitSize, Shift, &Done);
   {
     // Extract the last bit shifted in to CF
     const auto BitSize = IR::OpSizeToSize(Op->Size) * 8;
@@ -701,7 +701,7 @@ DEF_OP(RotateFlags) {
       msr(ARMEmitter::SystemRegister::NZCV, TMP3);
     }
   }
-  (void)Bind(&Done);
+  Bind(&Done);
 }
 
 DEF_OP(Extr) {
@@ -767,14 +767,14 @@ DEF_OP(PDep) {
     // Now, they're copied, so we can start setting Dest (even if it overlaps with
     // one of them).  Handle early exit case
     mov(EmitSize, Dest, 0);
-    (void)cbz(EmitSize, OrigMask, &Done);
+    cbz(EmitSize, OrigMask, &Done);
 
     // Setup for first iteration
     neg(EmitSize, T0, Mask);
     and_(EmitSize, T0, T0, Mask);
 
     // Main loop
-    (void)Bind(&NextBit);
+    Bind(&NextBit);
     sbfx(EmitSize, T1, Input, 0, 1);
     eor(EmitSize, Mask, Mask, T0);
     and_(EmitSize, T0, T1, T0);
@@ -782,10 +782,10 @@ DEF_OP(PDep) {
     orr(EmitSize, Dest, Dest, T0);
     lsr(EmitSize, Input, Input, 1);
     and_(EmitSize, T0, Mask, T1);
-    (void)cbnz(EmitSize, T0, &NextBit);
+    cbnz(EmitSize, T0, &NextBit);
 
     // All done with nothing to do.
-    (void)Bind(&Done);
+    Bind(&Done);
   }
 }
 
@@ -821,27 +821,27 @@ DEF_OP(PExt) {
     ARMEmitter::BackwardLabel NextBit;
     ARMEmitter::ForwardLabel Done;
 
-    (void)cbz(EmitSize, Mask, &EarlyExit);
+    cbz(EmitSize, Mask, &EarlyExit);
     mov(EmitSize, MaskReg, Mask);
     mov(EmitSize, ValueReg, Input);
     mov(EmitSize, Dest, ARMEmitter::Reg::zr);
 
     // Main loop
-    (void)Bind(&NextBit);
-    (void)cbz(EmitSize, MaskReg, &Done);
+    Bind(&NextBit);
+    cbz(EmitSize, MaskReg, &Done);
     clz(EmitSize, BitReg, MaskReg);
     lslv(EmitSize, ValueReg, ValueReg, BitReg);
     lslv(EmitSize, MaskReg, MaskReg, BitReg);
     extr(EmitSize, Dest, Dest, ValueReg, OpSizeBitsM1);
     bfc(EmitSize, MaskReg, OpSizeBitsM1, 1);
-    (void)b(&NextBit);
+    b(&NextBit);
 
     // Early exit
-    (void)Bind(&EarlyExit);
+    Bind(&EarlyExit);
     mov(EmitSize, Dest, ARMEmitter::Reg::zr);
 
     // All done with nothing to do.
-    (void)Bind(&Done);
+    Bind(&Done);
   }
 }
 
@@ -909,7 +909,7 @@ DEF_OP(Div) {
     eor(EmitSize, TMP1, TMP1, Upper);
 
     // If the sign bit matches then the result is zero
-    (void)cbz(EmitSize, TMP1, &Only64Bit);
+    cbz(EmitSize, TMP1, &Only64Bit);
 
     // Long divide
     {
@@ -928,17 +928,17 @@ DEF_OP(Div) {
       mov(EmitSize, Remainder, TMP2);
 
       // Skip 64-bit path
-      (void)b(&LongDIVRet);
+      b(&LongDIVRet);
     }
 
-    (void)Bind(&Only64Bit);
+    Bind(&Only64Bit);
     // 64-Bit only
     {
       sdiv(EmitSize, Quotient, Lower, Divisor);
       msub(EmitSize, Remainder, Quotient, Divisor, Lower);
     }
 
-    (void)Bind(&LongDIVRet);
+    Bind(&LongDIVRet);
     break;
   }
   default: LOGMAN_MSG_A_FMT("Unknown DIV Size: {}", OpSize); break;
@@ -992,7 +992,7 @@ DEF_OP(UDiv) {
 
     // Check the upper bits for zero
     // If the upper bits are zero then we can do a 64-bit divide
-    (void)cbz(EmitSize, Upper, &Only64Bit);
+    cbz(EmitSize, Upper, &Only64Bit);
 
     // Long divide
     {
@@ -1011,17 +1011,17 @@ DEF_OP(UDiv) {
       mov(EmitSize, Remainder, TMP2);
 
       // Skip 64-bit path
-      (void)b(&LongDIVRet);
+      b(&LongDIVRet);
     }
 
-    (void)Bind(&Only64Bit);
+    Bind(&Only64Bit);
     // 64-Bit only
     {
       udiv(EmitSize, Quotient, Lower, Divisor);
       msub(EmitSize, Remainder, Quotient, Divisor, Lower);
     }
 
-    (void)Bind(&LongDIVRet);
+    Bind(&LongDIVRet);
     break;
   }
   default: LOGMAN_MSG_A_FMT("Unknown LUDIV Size: {}", OpSize); break;
@@ -1190,6 +1190,19 @@ 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 +1286,16 @@ 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

@@ -63,7 +63,7 @@ void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair& Lit) {
   auto CurrentCursor = GetCursorAddress<uint8_t*>();
   Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - CodeData.BlockBegin;
 
-  BindOrRestart(&Lit.Loc);
+  Bind(&Lit.Loc);
   dc64(Lit.Lit);
   Relocations.emplace_back(Lit.MoveABI);
 }

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

@@ -62,27 +62,27 @@ DEF_OP(CASPair) {
     ARMEmitter::BackwardLabel LoopTop;
     ARMEmitter::ForwardLabel LoopNotExpected;
     ARMEmitter::ForwardLabel LoopExpected;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
 
     // This instruction sequence must be synced with HandleCASPAL_Armv8.
     ldaxp(EmitSize, TMP2, TMP3, MemSrc);
     cmp(EmitSize, TMP2, Expected0);
     ccmp(EmitSize, TMP3, Expected1, ARMEmitter::StatusFlags::None, ARMEmitter::Condition::CC_EQ);
-    (void)b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
+    b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
     stlxp(EmitSize, TMP2, Desired0, Desired1, MemSrc);
-    (void)cbnz(EmitSize, TMP2, &LoopTop);
+    cbnz(EmitSize, TMP2, &LoopTop);
     mov(EmitSize, Dst0, Expected0);
     mov(EmitSize, Dst1, Expected1);
 
-    (void)b(&LoopExpected);
+    b(&LoopExpected);
 
-    (void)Bind(&LoopNotExpected);
+    Bind(&LoopNotExpected);
     mov(EmitSize, Dst0, TMP2.R());
     mov(EmitSize, Dst1, TMP3.R());
     // exclusive monitor needs to be cleared here
     // Might have hit the case where ldaxr was hit but stlxr wasn't
     clrex();
-    (void)Bind(&LoopExpected);
+    Bind(&LoopExpected);
 
     // Restore
     msr(ARMEmitter::SystemRegister::NZCV, TMP1);
@@ -114,7 +114,7 @@ DEF_OP(CAS) {
     ARMEmitter::BackwardLabel LoopTop;
     ARMEmitter::ForwardLabel LoopNotExpected;
     ARMEmitter::ForwardLabel LoopExpected;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     if (IROp->Size == IR::OpSize::i8Bit) {
       cmp(EmitSize, TMP2, Expected, ARMEmitter::ExtendedType::UXTB, 0);
@@ -123,18 +123,18 @@ DEF_OP(CAS) {
     } else {
       cmp(EmitSize, TMP2, Expected);
     }
-    (void)b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
+    b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
     stlxr(SubEmitSize, TMP3, Desired, MemSrc);
-    (void)cbnz(EmitSize, TMP3, &LoopTop);
+    cbnz(EmitSize, TMP3, &LoopTop);
     mov(EmitSize, Dst, Expected);
-    (void)b(&LoopExpected);
+    b(&LoopExpected);
 
-    (void)Bind(&LoopNotExpected);
+    Bind(&LoopNotExpected);
     mov(EmitSize, Dst, TMP2.R());
     // exclusive monitor needs to be cleared here
     // Might have hit the case where ldaxr was hit but stlxr wasn't
     clrex();
-    (void)Bind(&LoopExpected);
+    Bind(&LoopExpected);
   }
 }
 
@@ -150,11 +150,11 @@ DEF_OP(AtomicXor) {
     steorl(SubEmitSize, Src, MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     eor(EmitSize, TMP2, TMP2, Src);
     stlxr(SubEmitSize, TMP2, TMP2, MemSrc);
-    (void)cbnz(EmitSize, TMP2, &LoopTop);
+    cbnz(EmitSize, TMP2, &LoopTop);
   }
 }
 
@@ -179,10 +179,10 @@ DEF_OP(AtomicSwap) {
     ldswpal(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     stlxr(SubEmitSize, TMP4, Src, MemSrc);
-    (void)cbnz(EmitSize, TMP4, &LoopTop);
+    cbnz(EmitSize, TMP4, &LoopTop);
     ubfm(EmitSize, GetReg(Node), TMP2, 0, IR::OpSizeAsBits(OpSize) - 1);
   }
 }
@@ -199,11 +199,11 @@ DEF_OP(AtomicFetchAdd) {
     ldaddal(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     add(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    (void)cbnz(EmitSize, TMP4, &LoopTop);
+    cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -221,11 +221,11 @@ DEF_OP(AtomicFetchSub) {
     ldaddal(SubEmitSize, TMP2, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     sub(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    (void)cbnz(EmitSize, TMP4, &LoopTop);
+    cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -243,11 +243,11 @@ DEF_OP(AtomicFetchAnd) {
     ldclral(SubEmitSize, TMP2, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     and_(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    (void)cbnz(EmitSize, TMP4, &LoopTop);
+    cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -264,11 +264,11 @@ DEF_OP(AtomicFetchCLR) {
     ldclral(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     bic(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    (void)cbnz(EmitSize, TMP4, &LoopTop);
+    cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -285,11 +285,11 @@ DEF_OP(AtomicFetchOr) {
     ldsetal(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     orr(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    (void)cbnz(EmitSize, TMP4, &LoopTop);
+    cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -306,11 +306,11 @@ DEF_OP(AtomicFetchXor) {
     ldeoral(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     eor(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    (void)cbnz(EmitSize, TMP4, &LoopTop);
+    cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -326,20 +326,20 @@ DEF_OP(AtomicFetchNeg) {
     // Use a CAS loop to avoid needing to emulate unaligned LLSC atomics
     ldr(SubEmitSize, TMP2, MemSrc);
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     mov(EmitSize, TMP4, TMP2);
     neg(EmitSize, TMP3, TMP2);
     casal(SubEmitSize, TMP2, TMP3, MemSrc);
     sub(EmitSize, TMP3, TMP2, TMP4);
-    (void)cbnz(EmitSize, TMP3, &LoopTop);
+    cbnz(EmitSize, TMP3, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     neg(EmitSize, TMP3, TMP2);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    (void)cbnz(EmitSize, TMP4, &LoopTop);
+    cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -359,11 +359,11 @@ DEF_OP(TelemetrySetValue) {
     stsetl(ARMEmitter::SubRegSize::i64Bit, TMP1, TMP2);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
+    Bind(&LoopTop);
     ldaxr(ARMEmitter::SubRegSize::i64Bit, TMP3, TMP2);
     orr(ARMEmitter::Size::i32Bit, TMP3, TMP3, Src);
     stlxr(ARMEmitter::SubRegSize::i64Bit, TMP3, TMP3, TMP2);
-    (void)cbnz(ARMEmitter::Size::i32Bit, TMP3, &LoopTop);
+    cbnz(ARMEmitter::Size::i32Bit, TMP3, &LoopTop);
   }
 #endif
 }

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

@@ -141,7 +141,7 @@ DEF_OP(ExitFunction) {
         if (!Op->CallReturnBlock.IsInvalid()) {
           auto CallReturnAddressReg = GetReg(Op->CallReturnAddress).X();
           PendingCallReturnTargetLabel = &CallReturnTargets.try_emplace(Op->CallReturnBlock.ID()).first->second;
-          (void)adr(TMP1, &l_CallReturn);
+          adr(TMP1, &l_CallReturn);
           stp<ARMEmitter::IndexType::PRE>(CallReturnAddressReg, TMP1, REG_CALLRET_SP, -0x10);
         } else {
           stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
@@ -149,16 +149,16 @@ DEF_OP(ExitFunction) {
       } else if (Op->Hint == IR::BranchHint::CheckTF) {
         ARMEmitter::ForwardLabel TFUnset;
         ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-        (void)cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
+        cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
         LoadConstant(ARMEmitter::Size::i64Bit, TMP1, NewRIP);
         str(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
         ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
         blr(TMP2);
-        (void)Bind(&TFUnset);
+        Bind(&TFUnset);
       }
 
       EmitLinkedBranch(NewRIP, Op->Hint == IR::BranchHint::Call);
-      (void)Bind(&l_CallReturn);
+      Bind(&l_CallReturn);
 #ifdef _M_ARM_64EC
     }
 #endif
@@ -170,7 +170,7 @@ DEF_OP(ExitFunction) {
       // First try to pop from the call-ret stack, otherwise follow the normal path (but ending in a ret)
       ldp<ARMEmitter::IndexType::POST>(TMP1, TMP2, REG_CALLRET_SP, 0x10);
       sub(TMP1, TMP1, RipReg.X());
-      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
+      cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
     }
 
     // L1 Cache
@@ -187,23 +187,23 @@ DEF_OP(ExitFunction) {
 
     // Note: sub+cbnz used over cmp+br to preserve flags.
     sub(TMP1, TMP1, RipReg.X());
-    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
+    cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
     ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
     str(RipReg.X(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
 
-    (void)Bind(&SkipFullLookup);
+    Bind(&SkipFullLookup);
     if (Op->Hint == IR::BranchHint::Call) {
       ARMEmitter::ForwardLabel l_CallReturn;
       if (!Op->CallReturnBlock.IsInvalid()) {
         auto CallReturnAddressReg = GetReg(Op->CallReturnAddress).X();
         PendingCallReturnTargetLabel = &CallReturnTargets.try_emplace(Op->CallReturnBlock.ID()).first->second;
-        (void)adr(TMP1, &l_CallReturn);
+        adr(TMP1, &l_CallReturn);
         stp<ARMEmitter::IndexType::PRE>(CallReturnAddressReg, TMP1, REG_CALLRET_SP, -0x10);
       } else {
         stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
       }
       blr(TMP2);
-      (void)Bind(&l_CallReturn);
+      Bind(&l_CallReturn);
     } else if (Op->Hint == IR::BranchHint::Return) {
       ret(TMP2);
     } else {
@@ -224,7 +224,7 @@ DEF_OP(CondJump) {
   auto TrueTargetLabel = JumpTarget(Op->TrueBlock);
 
   if (Op->FromNZCV) {
-    b_OrRestart(MapCC(Op->Cond), TrueTargetLabel);
+    b(MapCC(Op->Cond), TrueTargetLabel);
   } else {
     uint64_t Const;
     const bool isConst = IsInlineConstant(Op->Cmp2, &Const);
@@ -235,18 +235,18 @@ 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) {
+      cbz(Size, Reg, TrueTargetLabel);
+    } 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) {
+      cbnz(Size, Reg, TrueTargetLabel);
+    } 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) {
+      tbz(Reg, Const, TrueTargetLabel);
+    } else if (Op->Cond == IR::CondClass::TSTNZ) {
       LOGMAN_THROW_A_FMT(Const < 64, "CondJump: Expected valid bit source");
-      tbnz_OrRestart(Reg, Const, TrueTargetLabel);
+      tbnz(Reg, Const, TrueTargetLabel);
     } else {
       LOGMAN_THROW_A_FMT(false, "CondJump expected simple condition");
     }
@@ -458,7 +458,7 @@ DEF_OP(ValidateCode) {
     while (len >= Size) {
       LoadData();
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, TMP2);
-      cbnz_OrRestart(ARMEmitter::Size::i64Bit, TMP1, &Fail);
+      cbnz(ARMEmitter::Size::i64Bit, TMP1, &Fail);
       len -= Size;
       Offset += Size;
     }
@@ -486,10 +486,10 @@ DEF_OP(ValidateCode) {
 
   ARMEmitter::ForwardLabel End;
   LoadConstant(ARMEmitter::Size::i32Bit, Dst, 0);
-  b_OrRestart(&End);
-  BindOrRestart(&Fail);
+  b(&End);
+  Bind(&Fail);
   LoadConstant(ARMEmitter::Size::i32Bit, Dst, 1);
-  BindOrRestart(&End);
+  Bind(&End);
 }
 
 DEF_OP(ThreadRemoveCodeEntry) {

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

@@ -11,6 +11,8 @@ desc: Main glue logic of the arm64 splatter backend
 $end_info$
 */
 
+#include "Common/SoftFloat.h"
+
 #include "Interface/Context/Context.h"
 #include "Interface/Core/LookupCache.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
@@ -28,7 +30,6 @@ $end_info$
 #include <FEXCore/Utils/CompilerDefs.h>
 #include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/LogManager.h>
-#include <FEXCore/Utils/LongJump.h>
 #include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/Utils/Telemetry.h>
 #include <FEXCore/Utils/TypeDefines.h>
@@ -36,6 +37,7 @@ $end_info$
 
 #include <cstdio>
 #include <cstring>
+#include <limits>
 #include <unistd.h>
 
 namespace {
@@ -623,10 +625,10 @@ 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;
 
   {
@@ -667,6 +669,15 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::ContextImpl* ctx, FEXCore::Core::In
 
   CurrentCodeBuffer = CodeBuffers.GetLatest();
   ThreadState->LookupCache->Shared = CurrentCodeBuffer->LookupCache.get();
+
+  // Setup dynamic dispatch.
+  if (ParanoidTSO()) {
+    RT_LoadMemTSO = &Arm64JITCore::Op_ParanoidLoadMemTSO;
+    RT_StoreMemTSO = &Arm64JITCore::Op_ParanoidStoreMemTSO;
+  } else {
+    RT_LoadMemTSO = &Arm64JITCore::Op_LoadMemTSO;
+    RT_StoreMemTSO = &Arm64JITCore::Op_StoreMemTSO;
+  }
 }
 
 void Arm64JITCore::EmitDetectionString() {
@@ -737,11 +748,11 @@ void Arm64JITCore::EmitTFCheck() {
   // Note that this needs to be before the below suspend checks, as X86 checks this flag immediately after executing an instruction.
   ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
 
-  (void)cbz(ARMEmitter::Size::i32Bit, TMP1, &l_TFUnset);
+  cbz(ARMEmitter::Size::i32Bit, TMP1, &l_TFUnset);
 
   // X86 semantically checks TF after executing each instruction, so e.g. setting a context with TF set will execute a single instruction
   // and then raise an exception. However on the FEX side this is simpler to implement by checking at the start of each instruction, handle this by having bit 1 being unset in the flag state indicate that TF is blocked for a single instruction.
-  (void)tbz(TMP1, 1, &l_TFBlocked);
+  tbz(TMP1, 1, &l_TFBlocked);
 
   // Block TF for a single instruction when the frontend jumps to a new context by unsetting bit 1.
   ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
@@ -764,11 +775,11 @@ void Arm64JITCore::EmitTFCheck() {
   ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
   br(TMP1);
 
-  (void)Bind(&l_TFBlocked);
+  Bind(&l_TFBlocked);
   // If TF was blocked for this instruction, unblock it for the next.
   LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 0b11);
   strb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-  (void)Bind(&l_TFUnset);
+  Bind(&l_TFUnset);
 }
 
 void Arm64JITCore::EmitSuspendInterruptCheck() {
@@ -786,14 +797,14 @@ void Arm64JITCore::EmitSuspendInterruptCheck() {
   ARMEmitter::ForwardLabel l_NoSuspend;
   cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
   brk(SuspendMagic);
-  (void)Bind(&l_NoSuspend);
+  Bind(&l_NoSuspend);
 #endif
 }
 
 void Arm64JITCore::EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool CheckTF) {
   // Get the address of the JITCodeHeader and store in to the core state.
   // Two instruction cost, each 1 cycle.
-  adr_OrRestart(TMP1, &HeaderLabel);
+  adr(TMP1, &HeaderLabel);
   str(TMP1, STATE, offsetof(FEXCore::Core::CPUState, InlineJITBlockHeader));
 
   if (CheckTF) {
@@ -815,25 +826,16 @@ void Arm64JITCore::EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool C
 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");
-  this->Entry = Entry;
-  this->DebugData = DebugData;
-  this->IR = IR;
-  RequiresFarARM64Jumps = false;
 
-  switch (static_cast<RestartOptions::Control>(FEXCore::LongJump::SetJump(RestartControl.RestartJump))) {
-  case RestartOptions::Control::Incoming:
-    // Nothing
-    break;
-  case RestartOptions::Control::EnableFarARM64Jumps: RequiresFarARM64Jumps = true; break;
-  default: ERROR_AND_DIE_FMT("Unhandled Arm64 restart condition!");
-  }
-
-  uint32_t SSACount = IR->GetSSACount();
   JumpTargets.clear();
   CallReturnTargets.clear();
   PendingJumpThunks.clear();
+  uint32_t SSACount = IR->GetSSACount();
   JumpTargets.resize(IR->GetHeader()->BlockCount, {});
 
+  this->Entry = Entry;
+  this->DebugData = DebugData;
+  this->IR = IR;
   CodeData.EntryPoints.clear();
 
   // Fairly excessive buffer range to make sure we don't overflow
@@ -848,7 +850,7 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
   // Put the code header at the start of the data block.
   ARMEmitter::BackwardLabel JITCodeHeaderLabel {};
-  (void)Bind(&JITCodeHeaderLabel);
+  Bind(&JITCodeHeaderLabel);
   JITCodeHeader* CodeHeader = GetCursorAddress<JITCodeHeader*>();
   CursorIncrement(sizeof(JITCodeHeader));
 
@@ -896,7 +898,7 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
         if (PendingTargetLabel->Backward.Location) {
           EmitSuspendInterruptCheck();
         }
-        b_OrRestart(PendingTargetLabel);
+        b(PendingTargetLabel);
         PendingTargetLabel = nullptr;
       }
 
@@ -906,14 +908,14 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
         const auto IsReturnTarget = CallReturnTargets.try_emplace(Node).first;
         if (PendingTargetLabel) {
           // If there is a fallthrough branch to this block, skip over the entrypoint code.
-          b_OrRestart(Target);
+          b(Target);
         } else if (PendingCallReturnTargetLabel && PendingCallReturnTargetLabel != &IsReturnTarget->second) {
           // If we just emitted a call, but the block we're now emitting is not the return block so don't fallthrough.
-          b_OrRestart(PendingCallReturnTargetLabel);
+          b(PendingCallReturnTargetLabel);
         }
         PendingCallReturnTargetLabel = nullptr;
 
-        BindOrRestart(&IsReturnTarget->second);
+        Bind(&IsReturnTarget->second);
         CodeData.EntryPoints.emplace(BlockStartRIP, GetCursorAddress<uint8_t*>());
         DebugData->GuestOpcodes.push_back({BlockIROp->GuestEntryOffset, GetCursorAddress<uint8_t*>() - CodeData.BlockBegin});
 
@@ -922,12 +924,12 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
       if (PendingCallReturnTargetLabel) {
         // If there is still a pending call return target, then the block we're emitting is not the return block so don't fallthrough.
-        b_OrRestart(PendingCallReturnTargetLabel);
+        b(PendingCallReturnTargetLabel);
         PendingCallReturnTargetLabel = nullptr;
       }
       PendingTargetLabel = nullptr;
 
-      BindOrRestart(Target);
+      Bind(Target);
     }
 
     for (auto [CodeNode, IROp] : IR->GetCode(BlockNode)) {
@@ -954,7 +956,7 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
     if (PendingTargetLabel->Backward.Location) {
       EmitSuspendInterruptCheck();
     }
-    b_OrRestart(PendingTargetLabel);
+    b(PendingTargetLabel);
   }
   PendingTargetLabel = nullptr;
 
@@ -965,21 +967,21 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
     ARMEmitter::ForwardLabel l_DoLink;
     uint64_t ThunkAddress = GetCursorAddress<uint64_t>();
-    BindOrRestart(&PendingJumpThunk.Label);
-    b_OrRestart(&l_DoLink);
+    Bind(&PendingJumpThunk.Label);
+    b(&l_DoLink);
     br(TMP1);
-    BindOrRestart(&l_DoLink);
+    Bind(&l_DoLink);
     ldr(TMP1, &l_ExitLink);
     blr(TMP1);
 
     // This is a ExitFunctionLinkData struct
-    BindOrRestart(&l_ExitLink);
+    Bind(&l_ExitLink);
     dc64(0);                                             // HostCode
     dc64(PendingJumpThunk.GuestRIP);                     // GuestRIP
     dc64(PendingJumpThunk.CallerAddress - ThunkAddress); // CallerOffset
   }
 
-  BindOrRestart(&l_ExitLink);
+  Bind(&l_ExitLink);
   dc64(ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker);
 
   // CodeSize not including the header or tail data.

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

@@ -10,32 +10,38 @@ $end_info$
 #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>
 
 #include <CodeEmitter/Emitter.h>
 
 #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 {
@@ -55,7 +61,6 @@ public:
 
 private:
   FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
-  FEX_CONFIG_OPT(HalfBarrierTSOEnabled, HALFBARRIERTSOENABLED);
 
   const bool HostSupportsSVE128 {};
   const bool HostSupportsSVE256 {};
@@ -63,19 +68,6 @@ private:
   const bool HostSupportsRPRES {};
   const bool HostSupportsAFP {};
 
-  struct RestartOptions {
-    FEXCore::LongJump::JumpBuf RestartJump;
-    enum class Control : uint64_t {
-      Incoming = 0,
-      EnableFarARM64Jumps = 1,
-    };
-  };
-
-  // FEXCore makes assumptions in the JIT about certain conditions being true.
-  // In the rare case when those assumptions are broken, FEX needs to safely restart the JIT.
-  RestartOptions RestartControl {};
-  bool RequiresFarARM64Jumps {};
-
   ARMEmitter::BiDirectionalLabel* PendingTargetLabel {};
   ARMEmitter::BiDirectionalLabel* PendingCallReturnTargetLabel {};
   FEXCore::Context::ContextImpl* CTX {};
@@ -105,11 +97,13 @@ 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 +122,13 @@ 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 +146,8 @@ 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 +164,7 @@ 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 +172,23 @@ 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 +200,105 @@ 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]]
@@ -330,199 +331,14 @@ private:
   void EmitLinkedBranch(uint64_t GuestRIP, bool Call) {
     PendingJumpThunks.push_back({GetCursorAddress<uint64_t>(), GuestRIP, {}});
     auto& Thunk = PendingJumpThunks.back();
-    BindOrRestart(&Thunk.Label);
+    Bind(&Thunk.Label);
     if (Call) {
-      bl_OrRestart(&Thunk.Label);
+      bl(&Thunk.Label);
     } else {
-      b_OrRestart(&Thunk.Label);
+      b(&Thunk.Label);
     }
   }
 
-  // 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>)
-  void bl_OrRestart(T* Label) {
-    if (bl(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
-      return;
-    }
-
-    // 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));
-  }
-
-  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>)
-  void b_OrRestart(T* Label) {
-    if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
-      return;
-    }
-
-    // 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));
-  }
-
-  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>)
-  void b_OrRestart(ARMEmitter::Condition Cond, T* Label) {
-    if (RequiresFarARM64Jumps) {
-      ARMEmitter::ForwardLabel Skip {};
-      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
-      (void)b(InvertCondition(Cond), &Skip);
-      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
-        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-      }
-
-      (void)Bind(&Skip);
-      return;
-    }
-
-    if (b(Cond, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
-      return;
-    }
-
-    FEXCore::LongJump::LongJump(RestartControl.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>)
-  void cbz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
-    if (RequiresFarARM64Jumps) {
-      ARMEmitter::ForwardLabel Skip {};
-      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
-      (void)cbnz(s, rt, &Skip);
-      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
-        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-      }
-
-      (void)Bind(&Skip);
-      return;
-    }
-
-    if (cbz(s, rt, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
-      return;
-    }
-
-    FEXCore::LongJump::LongJump(RestartControl.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>)
-  void cbnz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
-    if (RequiresFarARM64Jumps) {
-      ARMEmitter::ForwardLabel Skip {};
-      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
-      (void)cbz(s, rt, &Skip);
-      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
-        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-      }
-
-      (void)Bind(&Skip);
-      return;
-    }
-
-    if (cbnz(s, rt, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
-      return;
-    }
-
-    FEXCore::LongJump::LongJump(RestartControl.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>)
-  void tbz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
-    if (RequiresFarARM64Jumps) {
-      ARMEmitter::ForwardLabel Skip {};
-      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
-      (void)tbnz(rt, Bit, &Skip);
-      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
-        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-      }
-
-      (void)Bind(&Skip);
-      return;
-    }
-
-    if (tbz(rt, Bit, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
-      return;
-    }
-
-    FEXCore::LongJump::LongJump(RestartControl.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>)
-  void tbnz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
-    if (RequiresFarARM64Jumps) {
-      ARMEmitter::ForwardLabel Skip {};
-      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
-      (void)tbz(rt, Bit, &Skip);
-      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
-        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-      }
-
-      (void)Bind(&Skip);
-      return;
-    }
-
-    if (tbnz(rt, Bit, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
-      return;
-    }
-
-    FEXCore::LongJump::LongJump(RestartControl.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>)
-  void adr_OrRestart(ARMEmitter::Register rd, T* Label) {
-    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));
-  }
-
-  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>)
-  void adrp_OrRestart(ARMEmitter::Register rd, T* Label) {
-    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));
-  }
-
-  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>)
-  void BindOrRestart(T* Label) {
-    if (Bind(Label)) {
-      return;
-    }
-
-    if (RequiresFarARM64Jumps) {
-      // This should have been caught before this point.
-      ERROR_AND_DIE_FMT("Oops. Unhandled long bind.");
-      return;
-    }
-
-    FEXCore::LongJump::LongJump(RestartControl.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
   void EmitDetectionString();
   IR::RegisterAllocationPass* RAPass {};
@@ -606,7 +422,7 @@ 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();
 
@@ -614,8 +430,17 @@ private:
 
   void EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool CheckTF);
 
+  // Runtime selection;
+  // Load and store TSO memory style
+  OpType RT_LoadMemTSO;
+  OpType RT_StoreMemTSO;
+
 #define DEF_OP(x) void Op_##x(IR::IROp_Header const* IROp, IR::Ref Node)
 
+  // Dynamic Dispatcher supporting operations
+  DEF_OP(ParanoidLoadMemTSO);
+  DEF_OP(ParanoidStoreMemTSO);
+
   ///< Unhandled handler
   DEF_OP(Unhandled);
 

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);
@@ -917,7 +912,7 @@ DEF_OP(VLoadVectorMasked) {
 
       // If the sign bit is zero then skip the load
       ARMEmitter::ForwardLabel Skip {};
-      (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+      tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
       // Do the gather load for this element into the destination
       switch (IROp->ElementSize) {
       case IR::OpSize::i8Bit: ld1<ARMEmitter::SubRegSize::i8Bit>(TempDst.Q(), i, TempMemReg); break;
@@ -928,7 +923,7 @@ DEF_OP(VLoadVectorMasked) {
       default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, IROp->ElementSize); return;
       }
 
-      (void)Bind(&Skip);
+      Bind(&Skip);
 
       if ((i + 1) != NumElements) {
         // Handle register rename to save a move.
@@ -1018,7 +1013,7 @@ DEF_OP(VStoreVectorMasked) {
 
       // If the sign bit is zero then skip the load
       ARMEmitter::ForwardLabel Skip {};
-      (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+      tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
       // Do the gather load for this element into the destination
       switch (IROp->ElementSize) {
       case IR::OpSize::i8Bit: st1<ARMEmitter::SubRegSize::i8Bit>(RegData.Q(), i, TempMemReg); break;
@@ -1029,7 +1024,7 @@ DEF_OP(VStoreVectorMasked) {
       default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, IROp->ElementSize); return;
       }
 
-      (void)Bind(&Skip);
+      Bind(&Skip);
 
       if ((i + 1) != NumElements) {
         // Handle register rename to save a move.
@@ -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) {
@@ -1107,7 +1102,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
     PerformMove(ElementSize, WorkingReg, MaskReg, i);
 
     // Skip if the mask's sign bit isn't set
-    (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+    tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
 
     // Extract Index Element
     if ((IndexElement * IR::OpSizeToSize(VectorIndexSize)) >= 16) {
@@ -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));
       }
     }
 
@@ -1145,7 +1140,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
     default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, ElementSize); FEX_UNREACHABLE;
     }
 
-    (void)Bind(&Skip);
+    Bind(&Skip);
   }
 
   if (NeedsDestTmp) {
@@ -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;
@@ -1880,7 +1874,7 @@ DEF_OP(MemSet) {
 
   if (!DirectionIsInline) {
     // Backward or forwards implementation depends on flag
-    (void)tbnz(DirectionReg, 1, &BackwardImpl);
+    tbnz(DirectionReg, 1, &BackwardImpl);
   }
 
   auto MemStore = [this](auto Value, uint32_t OpSize, int32_t Size) {
@@ -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;
@@ -1930,7 +1922,7 @@ DEF_OP(MemSet) {
     ARMEmitter::ForwardLabel DoneInternal {};
 
     // Early exit if zero count.
-    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+    cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
     if (!IsAtomic) {
       ARMEmitter::ForwardLabel AgainInternal256Exit {};
@@ -1947,50 +1939,50 @@ DEF_OP(MemSet) {
       // Do this in two parts, to fallback to the byte by byte loop if size < 32, and to the
       // single copy loop if size < 64.
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
+      tbnz(TMP1, 63, &AgainInternal128Exit);
 
       // Fill VTMP2 with the set pattern
       dup(SubRegSize, VTMP2.Q(), Value);
 
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)tbnz(TMP1, 63, &AgainInternal256Exit);
+      tbnz(TMP1, 63, &AgainInternal256Exit);
 
-      (void)Bind(&AgainInternal256);
+      Bind(&AgainInternal256);
       stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
       stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      (void)tbz(TMP1, 63, &AgainInternal256);
+      tbz(TMP1, 63, &AgainInternal256);
 
-      (void)Bind(&AgainInternal256Exit);
+      Bind(&AgainInternal256Exit);
       add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
-      (void)Bind(&AgainInternal128);
+      tbnz(TMP1, 63, &AgainInternal128Exit);
+      Bind(&AgainInternal128);
       stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)tbz(TMP1, 63, &AgainInternal128);
+      tbz(TMP1, 63, &AgainInternal128);
 
-      (void)Bind(&AgainInternal128Exit);
+      Bind(&AgainInternal128Exit);
       add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
       if (Direction == -1) {
         add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
       }
     }
 
-    (void)Bind(&AgainInternal);
+    Bind(&AgainInternal);
     if (IsAtomic) {
       MemStoreTSO(Value, OpSize, SizeDirection);
     } else {
       MemStore(Value, OpSize, SizeDirection);
     }
     sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 1);
-    (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
+    cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
 
-    (void)Bind(&DoneInternal);
+    Bind(&DoneInternal);
 
     if (SizeDirection >= 0) {
       switch (OpSize) {
@@ -2020,12 +2012,12 @@ DEF_OP(MemSet) {
       EmitMemset(Direction);
 
       if (Direction == 1) {
-        (void)b(&Done);
-        (void)Bind(&BackwardImpl);
+        b(&Done);
+        Bind(&BackwardImpl);
       }
     }
 
-    (void)Bind(&Done);
+    Bind(&Done);
     // Destination already set to the final pointer.
   }
 }
@@ -2075,7 +2067,7 @@ DEF_OP(MemCpy) {
 
   if (!DirectionIsInline) {
     // Backward or forwards implementation depends on flag
-    (void)tbnz(DirectionReg, 1, &BackwardImpl);
+    tbnz(DirectionReg, 1, &BackwardImpl);
   }
 
   auto MemCpy = [this](uint32_t OpSize, int32_t Size) {
@@ -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;
@@ -2176,7 +2164,7 @@ DEF_OP(MemCpy) {
     ARMEmitter::ForwardLabel DoneInternal {};
 
     // Early exit if zero count.
-    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+    cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
     if (!IsAtomic) {
       ARMEmitter::ForwardLabel AbsPos {};
@@ -2186,11 +2174,11 @@ DEF_OP(MemCpy) {
       ARMEmitter::BackwardLabel AgainInternal256 {};
 
       sub(ARMEmitter::Size::i64Bit, TMP4, TMP2, TMP3);
-      (void)tbz(TMP4, 63, &AbsPos);
+      tbz(TMP4, 63, &AbsPos);
       neg(ARMEmitter::Size::i64Bit, TMP4, TMP4);
-      (void)Bind(&AbsPos);
+      Bind(&AbsPos);
       sub(ARMEmitter::Size::i64Bit, TMP4, TMP4, 32);
-      (void)tbnz(TMP4, 63, &AgainInternal);
+      tbnz(TMP4, 63, &AgainInternal);
 
       if (Direction == -1) {
         sub(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
@@ -2202,30 +2190,30 @@ DEF_OP(MemCpy) {
       // Do this in two parts, to fallback to the byte by byte loop if size < 32, and to the
       // single copy loop if size < 64.
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
+      tbnz(TMP1, 63, &AgainInternal128Exit);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)tbnz(TMP1, 63, &AgainInternal256Exit);
+      tbnz(TMP1, 63, &AgainInternal256Exit);
 
-      (void)Bind(&AgainInternal256);
+      Bind(&AgainInternal256);
       MemCpy(32, 32 * Direction);
       MemCpy(32, 32 * Direction);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      (void)tbz(TMP1, 63, &AgainInternal256);
+      tbz(TMP1, 63, &AgainInternal256);
 
-      (void)Bind(&AgainInternal256Exit);
+      Bind(&AgainInternal256Exit);
       add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
-      (void)Bind(&AgainInternal128);
+      tbnz(TMP1, 63, &AgainInternal128Exit);
+      Bind(&AgainInternal128);
       MemCpy(32, 32 * Direction);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)tbz(TMP1, 63, &AgainInternal128);
+      tbz(TMP1, 63, &AgainInternal128);
 
-      (void)Bind(&AgainInternal128Exit);
+      Bind(&AgainInternal128Exit);
       add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
       if (Direction == -1) {
         add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
@@ -2233,16 +2221,16 @@ DEF_OP(MemCpy) {
       }
     }
 
-    (void)Bind(&AgainInternal);
+    Bind(&AgainInternal);
     if (IsAtomic) {
       MemCpyTSO(OpSize, SizeDirection);
     } else {
       MemCpy(OpSize, SizeDirection);
     }
     sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 1);
-    (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
+    cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
 
-    (void)Bind(&DoneInternal);
+    Bind(&DoneInternal);
 
     // Needs to use temporaries just in case of overwrite
     mov(TMP1, MemRegDest.X());
@@ -2300,15 +2288,186 @@ DEF_OP(MemCpy) {
     for (int32_t Direction : {1, -1}) {
       EmitMemcpy(Direction);
       if (Direction == 1) {
-        (void)b(&Done);
-        (void)Bind(&BackwardImpl);
+        b(&Done);
+        Bind(&BackwardImpl);
       }
     }
-    (void)Bind(&Done);
+    Bind(&Done);
     // Destination already set to the final pointer.
   }
 }
 
+DEF_OP(ParanoidLoadMemTSO) {
+  const auto Op = IROp->C<IR::IROp_LoadMemTSO>();
+  const auto OpSize = IROp->Size;
+
+  auto MemReg = GetReg(Op->Addr);
+
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
+    const auto Dst = GetReg(Node);
+    uint64_t Offset = 0;
+    if (!Op->Offset.IsInvalid()) {
+      if (!IsInlineConstant(Op->Offset, &Offset)) {
+        MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
+      }
+    }
+
+    if (OpSize == IR::OpSize::i8Bit) {
+      // 8bit load is always aligned to natural alignment
+      const auto Dst = GetReg(Node);
+      ldapurb(Dst, MemReg, Offset);
+    } else {
+      switch (OpSize) {
+      case IR::OpSize::i16Bit: ldapurh(Dst, MemReg, Offset); break;
+      case IR::OpSize::i32Bit: ldapur(Dst.W(), MemReg, Offset); break;
+      case IR::OpSize::i64Bit: ldapur(Dst.X(), MemReg, Offset); break;
+      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
+      }
+    }
+  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == IR::RegClass::GPR) {
+    const auto Dst = GetReg(Node);
+    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
+    if (OpSize == IR::OpSize::i8Bit) {
+      // 8bit load is always aligned to natural alignment
+      ldaprb(Dst.W(), MemReg);
+    } else {
+      switch (OpSize) {
+      case IR::OpSize::i16Bit: ldaprh(Dst.W(), MemReg); break;
+      case IR::OpSize::i32Bit: ldapr(Dst.W(), MemReg); break;
+      case IR::OpSize::i64Bit: ldapr(Dst.X(), MemReg); break;
+      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
+      }
+    }
+  } else if (Op->Class == IR::RegClass::GPR) {
+    const auto Dst = GetReg(Node);
+    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
+    switch (OpSize) {
+    case IR::OpSize::i8Bit: ldarb(Dst, MemReg); break;
+    case IR::OpSize::i16Bit: ldarh(Dst, MemReg); break;
+    case IR::OpSize::i32Bit: ldar(Dst.W(), MemReg); break;
+    case IR::OpSize::i64Bit: ldar(Dst.X(), MemReg); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
+    }
+  } else {
+    const auto Dst = GetVReg(Node);
+    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
+    switch (OpSize) {
+    case IR::OpSize::i8Bit:
+      ldarb(TMP1, MemReg);
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
+      break;
+    case IR::OpSize::i16Bit:
+      ldarh(TMP1, MemReg);
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
+      break;
+    case IR::OpSize::i32Bit:
+      ldar(TMP1.W(), MemReg);
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
+      break;
+    case IR::OpSize::i64Bit:
+      ldar(TMP1, MemReg);
+      fmov(ARMEmitter::Size::i64Bit, Dst.D(), TMP1);
+      break;
+    case IR::OpSize::i128Bit:
+      ldaxp(ARMEmitter::Size::i64Bit, TMP1, TMP2, MemReg);
+      clrex();
+      ins(ARMEmitter::SubRegSize::i64Bit, Dst, 0, TMP1);
+      ins(ARMEmitter::SubRegSize::i64Bit, Dst, 1, TMP2);
+      break;
+    case IR::OpSize::i256Bit:
+      LOGMAN_THROW_A_FMT(HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__);
+      dmb(ARMEmitter::BarrierScope::ISH);
+      ld1b<ARMEmitter::SubRegSize::i8Bit>(Dst.Z(), PRED_TMP_32B.Zeroing(), MemReg);
+      dmb(ARMEmitter::BarrierScope::ISH);
+      break;
+    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
+    }
+  }
+}
+
+DEF_OP(ParanoidStoreMemTSO) {
+  const auto Op = IROp->C<IR::IROp_StoreMemTSO>();
+  const auto OpSize = IROp->Size;
+
+  auto MemReg = GetReg(Op->Addr);
+
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
+    const auto Src = GetZeroableReg(Op->Value);
+    uint64_t Offset = 0;
+    if (!Op->Offset.IsInvalid()) {
+      if (!IsInlineConstant(Op->Offset, &Offset)) {
+        MemReg = ApplyMemOperand(OpSize, MemReg, TMP1, Op->Offset, Op->OffsetType, Op->OffsetScale);
+      }
+    }
+
+    if (OpSize == IR::OpSize::i8Bit) {
+      // 8bit load is always aligned to natural alignment
+      stlurb(Src, MemReg, Offset);
+    } else {
+      switch (OpSize) {
+      case IR::OpSize::i16Bit: stlurh(Src, MemReg, Offset); break;
+      case IR::OpSize::i32Bit: stlur(Src.W(), MemReg, Offset); break;
+      case IR::OpSize::i64Bit: stlur(Src.X(), MemReg, Offset); break;
+      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
+      }
+    }
+  } else if (Op->Class == IR::RegClass::GPR) {
+    const auto Src = GetZeroableReg(Op->Value);
+    MemReg = ApplyMemOperand(OpSize, MemReg, TMP1, Op->Offset, Op->OffsetType, Op->OffsetScale);
+    switch (OpSize) {
+    case IR::OpSize::i8Bit: stlrb(Src, MemReg); break;
+    case IR::OpSize::i16Bit: stlrh(Src, MemReg); break;
+    case IR::OpSize::i32Bit: stlr(Src.W(), MemReg); break;
+    case IR::OpSize::i64Bit: stlr(Src.X(), MemReg); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
+    }
+  } else {
+    const auto Src = GetVReg(Op->Value);
+
+    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
+
+    switch (OpSize) {
+    case IR::OpSize::i8Bit:
+      umov<ARMEmitter::SubRegSize::i8Bit>(TMP1, Src, 0);
+      stlrb(TMP1, MemReg);
+      break;
+    case IR::OpSize::i16Bit:
+      umov<ARMEmitter::SubRegSize::i16Bit>(TMP1, Src, 0);
+      stlrh(TMP1, MemReg);
+      break;
+    case IR::OpSize::i32Bit:
+      umov<ARMEmitter::SubRegSize::i32Bit>(TMP1, Src, 0);
+      stlr(TMP1.W(), MemReg);
+      break;
+    case IR::OpSize::i64Bit:
+      umov<ARMEmitter::SubRegSize::i64Bit>(TMP1, Src, 0);
+      stlr(TMP1, MemReg);
+      break;
+    case IR::OpSize::i128Bit: {
+      // Move vector to GPRs
+      umov<ARMEmitter::SubRegSize::i64Bit>(TMP1, Src, 0);
+      umov<ARMEmitter::SubRegSize::i64Bit>(TMP2, Src, 1);
+      ARMEmitter::BackwardLabel B;
+      Bind(&B);
+
+      // ldaxp must not have both the destination registers be the same
+      ldaxp(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::zr, TMP3, MemReg); // <- Can hit SIGBUS. Overwritten with DMB
+      stlxp(ARMEmitter::Size::i64Bit, TMP3, TMP1, TMP2, MemReg);          // <- Can also hit SIGBUS
+      cbnz(ARMEmitter::Size::i64Bit, TMP3, &B);                           // < Overwritten with DMB
+      break;
+    }
+    case IR::OpSize::i256Bit: {
+      LOGMAN_THROW_A_FMT(HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__);
+      dmb(ARMEmitter::BarrierScope::ISH);
+      st1b<ARMEmitter::SubRegSize::i8Bit>(Src.Z(), PRED_TMP_32B, MemReg, 0);
+      dmb(ARMEmitter::BarrierScope::ISH);
+      break;
+    }
+    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
+    }
+  }
+}
+
 DEF_OP(CacheLineClear) {
   if (!CTX->HostFeatures.SupportsCacheMaintenanceOps) {
     dmb(ARMEmitter::BarrierScope::SY);

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/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/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);
           }
         }
       }
@@ -1329,6 +1330,7 @@ protected:
 
 private:
   FEX_CONFIG_OPT(ReducedPrecisionMode, X87REDUCEDPRECISION);
+  FEX_CONFIG_OPT(StrictReducedPrecisionMode, X87STRICTREDUCEDPRECISION);
 
   struct JumpTargetInfo {
     Ref BlockEntry;
@@ -1553,23 +1555,63 @@ 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]]
@@ -1805,14 +1847,15 @@ 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 +1881,12 @@ 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 +1917,11 @@ 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 +1953,14 @@ 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 +1974,7 @@ 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 +1985,7 @@ 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 +1994,21 @@ 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 +2016,7 @@ 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 +2028,13 @@ 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 +2050,7 @@ 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 +2105,7 @@ 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 +2122,18 @@ 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 +2222,7 @@ 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 +2295,7 @@ 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 +2309,8 @@ 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 +2345,7 @@ 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 +2388,7 @@ 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 +2405,7 @@ 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 +2442,62 @@ 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 +2515,72 @@ 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),
-      };
+      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);
+      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);
@@ -1968,7 +1964,7 @@ 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], SrcSize, Op->Flags);
   }
 
   Ref Sources[3] = {Dest, Src1, Src2};
@@ -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

@@ -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

@@ -28,7 +28,7 @@ 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) {
@@ -52,18 +52,18 @@ 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) {
@@ -79,14 +79,14 @@ 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);
 }
 
 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();
 }
 
@@ -99,7 +99,7 @@ void OpDispatchBuilder::FLD_Const(OpcodeArgs, NamedVectorConstant K) {
 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) {
@@ -112,15 +112,15 @@ 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);
@@ -166,12 +166,12 @@ 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 +180,7 @@ 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 +206,10 @@ 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 +236,10 @@ 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 +273,10 @@ 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 +314,10 @@ 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 +340,7 @@ 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 +381,41 @@ 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 +441,20 @@ 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 +483,61 @@ 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 +548,8 @@ 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 +561,11 @@ 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 +574,14 @@ 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 +590,19 @@ 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 +610,8 @@ 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 +647,10 @@ 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 +766,7 @@ 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 +785,12 @@ 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 +863,7 @@ 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

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

@@ -29,38 +29,38 @@ 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) {
@@ -73,7 +73,7 @@ void OpDispatchBuilder::FLDF64(OpcodeArgs, IR::OpSize 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);
@@ -82,7 +82,7 @@ void OpDispatchBuilder::FBLDF64(OpcodeArgs) {
 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();
 }
 
@@ -95,7 +95,7 @@ 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);
   }
@@ -112,7 +112,7 @@ 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 +138,16 @@ 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 +176,16 @@ 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 +228,16 @@ 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 +285,16 @@ 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 +332,16 @@ 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;
@@ -393,8 +393,8 @@ 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);

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

@@ -442,7 +442,7 @@ 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
@@ -431,94 +430,6 @@ 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 +693,15 @@ 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 +710,3 @@ 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 +136,13 @@
     "GPR": "OrderedNode*",
     "FPR": "OrderedNode*",
     "FenceType": "FenceType",
-    "RegisterClass": "RegisterClassType",
-    "CondClass": "CondClassType",
+    "RegisterClass": "RegClass",
+    "CondClass": "CondClass",
     "SyscallFlags": "FEXCore::IR::SyscallFlags",
     "SHA256Sum": "SHA256Sum",
     "MemOffsetType": "MemOffsetType",
     "BreakDefinition": "BreakDefinition",
-    "RoundType": "RoundType",
+    "RoundType": "RoundMode",
     "FloatCompareOp": "FloatCompareOp",
     "NamedVectorConstant": "FEXCore::IR::NamedVectorConstant",
     "IndexNamedVectorConstant": "FEXCore::IR::IndexNamedVectorConstant",
@@ -307,7 +295,7 @@
         "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"
@@ -364,20 +352,6 @@
       "GPR = Copy GPR:$Source": {
         "Desc": ["GPR copy, generated by RA to split live ranges"],
         "DestSize": "OpSize::i64Bit"
-      },
-
-      "GPR = Swap1 GPR:$A, GPR:$B": {
-        "Desc": ["GPR swap part 1, generated by RA. Returns value of first source.",
-                 "Destination must be second GPR."],
-        "DestSize": "OpSize::i64Bit"
-      },
-
-      "GPR = Swap2": {
-        "Desc": ["GPR swap part 2, generated by RA. Returns source source.",
-                 "Must immediately succeed Swap1 with no intervening instructions",
-                 "Kludge to workaround single destination restriction on IR",
-                 "Hopefully temporary"],
-        "DestSize": "OpSize::i64Bit"
       }
     },
     "StaticRA": {
@@ -423,8 +397,8 @@
                 ],
         "DestSize": "ByteSize",
         "EmitValidation": [
-          "($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]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't LoadContext to XMM\""
         ]
@@ -437,8 +411,8 @@
         "HasSideEffects": true,
         "DestSize": "ByteSize",
         "EmitValidation": [
-          "($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]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't LoadContext to XMM\""
         ]
@@ -454,8 +428,8 @@
         "HasSideEffects": true,
         "DestSize": "ByteSize",
         "EmitValidation": [
-          "($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]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't StoreContext to XMM\""
         ]
@@ -472,8 +446,8 @@
         "EmitValidation": [
           "WalkFindRegClass($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]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't StoreContext to XMM\""
         ]
@@ -485,8 +459,8 @@
                 ],
         "DestSize": "ByteSize",
         "EmitValidation": [
-          "($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]) && $BaseOffset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't LoadContextIndexed to XMM\""
         ]
@@ -499,8 +473,8 @@
         "DestSize": "ByteSize",
         "EmitValidation": [
           "WalkFindRegClass($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\""
         ]
@@ -604,7 +578,8 @@
         "Desc": ["Does a x86 TSO compatible load from memory. Offset must be Invalid()."
                 ],
         "Inline": ["", "Memtso"],
-        "DestSize": "Size"
+        "DestSize": "Size",
+        "DynamicDispatch": true
       },
 
       "StoreMemTSO RegisterClass:$Class, OpSize:#Size, SSA:$Value, GPR:$Addr, GPR:$Offset, OpSize:$Align, MemOffsetType:$OffsetType, u8:$OffsetScale": {
@@ -612,7 +587,8 @@
                 ],
         "Inline": ["Zero", "", "Memtso"],
         "HasSideEffects": true,
-        "DestSize": "Size"
+        "DestSize": "Size",
+        "DynamicDispatch": true
       },
 
       "FPR = VLoadVectorMasked OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Mask, GPR:$Addr, GPR:$Offset, MemOffsetType:$OffsetType, u8:$OffsetScale": {
@@ -630,7 +606,7 @@
         "DestSize": "RegisterSize",
         "ElementSize": "ElementSize"
       },
-      "FPR = VLoadVectorGatherMasked OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$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 +620,7 @@
           "$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 +735,10 @@
       },
       "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
@@ -1006,7 +983,7 @@
         "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 +1061,13 @@
           "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:$Src2": {
         "Desc": [ "Integer Add",
                   "Will truncate to 64 or 32bits"
@@ -1551,6 +1535,15 @@
           "ResultSize == FEXCore::IR::OpSize::i32Bit || ResultSize == FEXCore::IR::OpSize::i64Bit"
         ]
       },
+      "GPR = MaskGenerateFromBitWidth GPR:$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 +2152,34 @@
 
       "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": {
@@ -2838,7 +2843,7 @@
           "Int: 64-bit, 32-bit, 16-bit"
         ],
         "EmitValidation": [
-          "WalkFindRegClass($OriginalValue) == FPRClass || WalkFindRegClass($OriginalValue) == GPRClass"
+          "WalkFindRegClass($OriginalValue) == RegClass::FPR || WalkFindRegClass($OriginalValue) == RegClass::GPR"
         ],
         "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 +124,46 @@ 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*, SyscallFlags Arg) {
+  *out << [Arg] {
+    switch (Arg) {
+    case SyscallFlags::DEFAULT: return "Default";
+    case SyscallFlags::OPTIMIZETHROUGH: return "Optimize Through";
+    case SyscallFlags::NOSYNCSTATEONENTRY: return "No Sync State on Entry";
+    case SyscallFlags::NORETURN: return "No Return";
+    case SyscallFlags::NOSIDEEFFECTS: return "No Side Effects";
+    case SyscallFlags::NORETURNEDRESULT: return "No Returned Result";
+    }
+    return "<Unknown Syscall Flags>";
+  }();
 }
 
-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 +191,22 @@ 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 +237,20 @@ 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 +273,43 @@ 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 +319,28 @@ 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) {
@@ -353,17 +397,17 @@ 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/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
@@ -56,11 +57,11 @@ 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;
@@ -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);
   };
 
@@ -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

@@ -12,11 +12,11 @@ $end_info$
 #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;

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

@@ -158,7 +158,9 @@ public:
     : Features(Features)
     , GPROpSize(GPROpSize) {
     FEX_CONFIG_OPT(ReducedPrecision, X87REDUCEDPRECISION);
+    FEX_CONFIG_OPT(StrictReducedPrecision, X87STRICTREDUCEDPRECISION);
     ReducedPrecisionMode = ReducedPrecision;
+    StrictReducedPrecisionMode = StrictReducedPrecision;
   }
   void Run(IREmitter* Emit) override;
 
@@ -166,10 +168,12 @@ private:
   const FEXCore::HostFeatures& Features;
   const OpSize GPROpSize;
   bool ReducedPrecisionMode;
+  bool StrictReducedPrecisionMode;
   FEX_CONFIG_OPT(DisableVixlIndirectCalls, DISABLE_VIXL_INDIRECT_RUNTIME_CALLS);
 
   // Helpers
   Ref RotateRight8(uint32_t V, Ref Amount);
+  Ref SilenceNaN(Ref Value);
 
   void F80SplitStore_Helper(const IROp_StoreStackMem* Op, Ref StackNode) {
     Ref AddrNode = IR->GetNode(Op->Addr);
@@ -178,18 +182,18 @@ private:
     MemOffsetType OffsetType = Op->OffsetType;
     uint8_t OffsetScale = Op->OffsetScale;
 
-    IREmit->_StoreMem(FPRClass, OpSize::i64Bit, StackNode, AddrNode, Offset, Align, OffsetType, 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 StoreStackMem_Helper(const IROp_StoreStackMem* Op, Ref StackNode) {
@@ -204,7 +208,10 @@ private:
     case OpSize::i32Bit:
     case OpSize::i64Bit: {
       StackNode = IREmit->_F80CVT(Op->StoreSize, StackNode);
-      IREmit->_StoreMem(FPRClass, Op->StoreSize, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
+      if (!ReducedPrecisionMode || StrictReducedPrecisionMode) {
+        StackNode = SilenceNaN(StackNode);
+      }
+      IREmit->_StoreMemFPR(Op->StoreSize, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
       break;
     }
 
@@ -212,7 +219,7 @@ private:
       if (Features.SupportsSVE128 || Features.SupportsSVE256) {
         AddressMode A {.Base = AddrNode,
                        .Index = Op->Offset.IsInvalid() ? nullptr : Offset,
-                       .IndexType = MEM_OFFSET_SXTX,
+                       .IndexType = MemOffsetType::SXTX,
                        .IndexScale = OffsetScale,
                        .AddrSize = OpSize::i64Bit};
         AddrNode = LoadEffectiveAddress(IREmit, A, GPROpSize, false);
@@ -235,13 +242,17 @@ private:
     MemOffsetType OffsetType = Op->OffsetType;
     uint8_t OffsetScale = Op->OffsetScale;
 
+    if ((!ReducedPrecisionMode || StrictReducedPrecisionMode) && Op->StoreSize != OpSize::f80Bit) {
+      StackNode = SilenceNaN(StackNode);
+    }
+
     switch (Op->StoreSize) {
     case OpSize::i32Bit: {
       StackNode = IREmit->_Float_FToF(OpSize::i32Bit, OpSize::i64Bit, StackNode);
       [[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;
     }
 
@@ -401,8 +412,7 @@ 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 +457,7 @@ 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 +480,7 @@ 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];
 }
@@ -489,6 +499,15 @@ inline Ref X87StackOptimization::RotateRight8(uint32_t V, Ref Amount) {
   return IREmit->_Lshr(OpSize::i32Bit, GetConstant(V | (V << 8)), Amount);
 }
 
+inline Ref X87StackOptimization::SilenceNaN(Ref Value) {
+  Ref GPRValue = IREmit->_VExtractToGPR(OpSize::i64Bit, OpSize::i64Bit, Value, 0);
+
+  IREmit->_FCmp(OpSize::i64Bit, Value, Value); // Comparison with itself should set VS if nan
+  Ref QuietNaNGPR = IREmit->_Or(OpSize::i64Bit, GPRValue, IREmit->_Constant(0x0008000000000000ULL));
+  Ref SilencedValue = IREmit->_VCastFromGPR(OpSize::i64Bit, OpSize::i64Bit, QuietNaNGPR);
+  return IREmit->_NZCVSelectV(OpSize::i64Bit, CondClass::VS, SilencedValue, Value);
+}
+
 inline std::optional<X87StackOptimization::StackMemberInfo> X87StackOptimization::MigrateToSlowPath_IfInvalid(uint8_t Offset) {
   const auto& [Valid, StackMember] = StackData.top(Offset);
   MigrateToSlowPathIf(Valid != StackSlot::VALID);
@@ -595,7 +614,7 @@ 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 +622,7 @@ 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 +673,7 @@ 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;
     }
 
@@ -995,8 +1014,8 @@ void X87StackOptimization::Run(IREmitter* Emit) {
         // 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);
+          const auto ClassType = Value->InterpretAsFloat ? RegClass::FPR : RegClass::GPR;
+          IREmit->_StoreMem(ClassType, Op->StoreSize, Value->Source->Node, AddrNode, Offset, Align, OffsetType, OffsetScale);
           break;
         }
 
@@ -1160,7 +1179,7 @@ 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>
@@ -50,6 +51,10 @@ 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) {
+    prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, Result, length, "FEXMem");
+  }
   return Result;
 }
 int FEX_munmap(void* addr, size_t length) {

FEXCore/Source/Utils/LongJump.cpp

@@ -1,119 +0,0 @@
-// SPDX-License-Identifier: MIT
-#include <FEXCore/Utils/LongJump.h>
-
-namespace FEXCore::LongJump {
-#if defined(_M_ARM_64)
-[[nodiscard]]
-FEX_DEFAULT_VISIBILITY FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
-  __asm volatile(R"(
-      // x0 contains the jumpbuffer
-      stp x19, x20, [x0, #( 0 * 8)];
-      stp x21, x22, [x0, #( 2 * 8)];
-      stp x23, x24, [x0, #( 4 * 8)];
-      stp x25, x26, [x0, #( 6 * 8)];
-      stp x27, x28, [x0, #( 8 * 8)];
-      stp x29, x30, [x0, #(10 * 8)];
-
-      // FPRs
-      stp d8,   d9, [x0, #(12 * 8)];
-      stp d10, d11, [x0, #(14 * 8)];
-      stp d12, d13, [x0, #(16 * 8)];
-      stp d14, d15, [x0, #(18 * 8)];
-
-      // Move SP in to a temporary to store.
-      mov x1, sp;
-      str x1,  [x0, #(20 * 8)];
-
-      // Return zero to signify this is the SetJump.
-      mov x0, #0;
-      ret;
-    )" ::
-                   : "memory");
-}
-
-[[noreturn]]
-FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value) {
-  __asm volatile(R"(
-      // x0 contains the jumpbuffer
-      ldp x19, x20, [x0, #( 0 * 8)];
-      ldp x21, x22, [x0, #( 2 * 8)];
-      ldp x23, x24, [x0, #( 4 * 8)];
-      ldp x25, x26, [x0, #( 6 * 8)];
-      ldp x27, x28, [x0, #( 8 * 8)];
-      ldp x29, x30, [x0, #(10 * 8)];
-
-      // FPRs
-      ldp d8,   d9, [x0, #(12 * 8)];
-      ldp d10, d11, [x0, #(14 * 8)];
-      ldp d12, d13, [x0, #(16 * 8)];
-      ldp d14, d15, [x0, #(18 * 8)];
-
-      // Load SP in to temporary then move
-      ldr x0,  [x0, #(20 * 8)];
-      mov sp, x0;
-
-      // Move value in to result register
-      mov x0, x1;
-      ret;
-    )" ::
-                   : "memory");
-}
-#else
-[[nodiscard]]
-FEX_DEFAULT_VISIBILITY FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
-  __asm volatile(R"(
-    .intel_syntax noprefix;
-    // rdi contains the jumpbuffer
-    mov [rdi + (0 * 8)], rbx;
-    mov [rdi + (1 * 8)], rsp;
-    mov [rdi + (2 * 8)], rbp;
-    mov [rdi + (3 * 8)], r12;
-    mov [rdi + (4 * 8)], r13;
-    mov [rdi + (5 * 8)], r14;
-    mov [rdi + (6 * 8)], r15;
-
-    // Return address is on the stack, load it and store
-    mov rsi, [rsp];
-    mov [rdi + (7 * 8)], rsi;
-
-    // Return zero to signify this is the SetJump.
-    mov rax, 0;
-    ret;
-
-    .att_syntax prefix;
-    )" ::
-                   : "memory");
-}
-
-[[noreturn]]
-FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value) {
-  __asm volatile(R"(
-    .intel_syntax noprefix;
-    // rdi contains the jumpbuffer
-    mov rbx, [rdi + (0 * 8)];
-    mov rsp, [rdi + (1 * 8)];
-    mov rbp, [rdi + (2 * 8)];
-    mov r12, [rdi + (3 * 8)];
-    mov r13, [rdi + (4 * 8)];
-    mov r14, [rdi + (5 * 8)];
-    mov r15, [rdi + (6 * 8)];
-
-    // Move value in to result register
-    mov rax, rsi;
-
-    // Pop the dead return address off the stack
-    pop rsi;
-
-    // Load the original return address from the jumpbuffer
-    mov rsi, [rdi + (7 * 8)];
-
-    // Return using a jump
-    jmp rsi;
-
-    .att_syntax prefix;
-    )" ::
-                   : "memory");
-}
-
-#endif
-} // namespace FEXCore::LongJump

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/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/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

@@ -1,36 +0,0 @@
-// SPDX-License-Identifier: MIT
-#pragma once
-#include <FEXCore/Utils/CompilerDefs.h>
-
-#include <cstdint>
-
-// It's longjump without glibc fortification checks.
-namespace FEXCore::LongJump {
-// JumpBuf definition needs to be public because the frontend needs to understand it.
-#if defined(_M_ARM_64)
-struct JumpBuf {
-  // All the registers that are required by AAPCS64 to save.
-  // GPRs
-  // X19, X20, X21, X22,
-  // X23, X24, X25, X26,
-  // X27, X28, X29, X30,
-  //
-  // Lower 64-bits:
-  //  V8,  V9, V10, V11,
-  // V12, V13, V14, V15,
-  //
-  // SP,
-  uint64_t Registers[21];
-};
-#else
-struct JumpBuf {
-  // Registers to preserve
-  // RBX, RSP, RBP, R12, R13, R14, R15,
-  // <return address>
-  uint64_t Registers[8];
-};
-#endif
-
-[[nodiscard]] FEX_DEFAULT_VISIBILITY uint64_t SetJump(JumpBuf& Buffer);
-[[noreturn]] FEX_DEFAULT_VISIBILITY void LongJump(JumpBuf& Buffer, uint64_t Value);
-} // namespace FEXCore::LongJump

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

FEXCore/include/FEXCore/fextl/string.h

@@ -2,7 +2,7 @@
 #pragma once
 #include <FEXCore/fextl/allocator.h>
 
-#include <filesystem>
+#include <functional>
 #include <string>
 
 namespace fextl {

FEXCore/unittests/Emitter/ALU_Tests.cpp

@@ -9,17 +9,17 @@ using namespace ARMEmitter;
 TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)adr(Reg::r30, &Label);
+    adr(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x10fffffe);
   }
 
   {
     ForwardLabel Label;
-    (void)adr(Reg::r30, &Label);
-    (void)Bind(&Label);
+    adr(Reg::r30, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x1000003e);
@@ -27,17 +27,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)adr(Reg::r30, &Label);
+    adr(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x10fffffe);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)adr(Reg::r30, &Label);
-    (void)Bind(&Label);
+    adr(Reg::r30, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x1000003e);
@@ -45,80 +45,80 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)adrp(Reg::r30, &Label);
+    adrp(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x9000001e);
   }
 
   {
     ForwardLabel Label;
-    (void)adrp(Reg::r30, &Label);
+    adrp(Reg::r30, &Label);
     // Move label a page away
     for (size_t i = 0; i < 1023; ++i) {
       nop();
     }
-    (void)Bind(&Label);
+    Bind(&Label);
 
     CHECK(DisassembleEncoding(0) == 0xb000001e);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)adrp(Reg::r30, &Label);
+    adrp(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x9000001e);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)adrp(Reg::r30, &Label);
+    adrp(Reg::r30, &Label);
     // Move label a page away
     for (size_t i = 0; i < 1023; ++i) {
       nop();
     }
-    (void)Bind(&Label);
+    Bind(&Label);
 
     CHECK(DisassembleEncoding(0) == 0xb000001e);
   }
 
   {
-    // Will generate (void)adr.
+    // Will generate adr.
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
-    (void)LongAddressGen(Reg::r30, &Label);
+    LongAddressGen(Reg::r30, &Label);
     CHECK(DisassembleEncoding(1) == 0x10fffffe);
   }
   {
-    // Will generate nop + (void)adr.
+    // Will generate nop + adr.
     ForwardLabel Label;
-    (void)LongAddressGen(Reg::r30, &Label);
-    (void)Bind(&Label);
+    LongAddressGen(Reg::r30, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd503201f);
     CHECK(DisassembleEncoding(1) == 0x1000003e);
   }
   {
-    // Will generate (void)adr.
+    // Will generate adr.
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)LongAddressGen(Reg::r30, &Label);
+    LongAddressGen(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x10fffffe);
   }
 
   {
-    // Will generate nop + (void)adr.
+    // Will generate nop + adr.
     BiDirectionalLabel Label;
-    (void)LongAddressGen(Reg::r30, &Label);
-    (void)Bind(&Label);
+    LongAddressGen(Reg::r30, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd503201f);
@@ -126,33 +126,33 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   }
 
   {
-    // Will generate (void)adrp.
+    // Will generate adrp.
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
-    // Move (void)adrp 1MB away.
+    // Move adrp 1MB away.
     for (size_t i = 0; i < (1 * 1024 * 1024 / 4); ++i) {
       nop();
     }
 
-    (void)LongAddressGen(Reg::r30, &Label);
+    LongAddressGen(Reg::r30, &Label);
     nop();
     CHECK(DisassembleEncoding(262145) == 0x90fff81e);
     CHECK(DisassembleEncoding(262146) == 0xd503201f);
   }
 
   {
-    // Will generate nop + (void)adrp.
+    // Will generate nop + adrp.
     ForwardLabel Label;
-    (void)LongAddressGen(Reg::r30, &Label);
+    LongAddressGen(Reg::r30, &Label);
 
     // Move label 1MB away, plus a page, and then aligned to a page.
     for (size_t i = 0; i < ((1 * 1024 * 1024 + 4096) / 4 - 2); ++i) {
       nop();
     }
 
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd503201f);
@@ -160,16 +160,16 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   }
 
   {
-    // Will generate (void)adrp + add.
+    // Will generate adrp + add.
     ForwardLabel Label;
-    (void)LongAddressGen(Reg::r30, &Label);
+    LongAddressGen(Reg::r30, &Label);
 
     // Move label 1MB away, plus a page, plus one instruction.
     for (size_t i = 0; i < ((1 * 1024 * 1024 + 4096) / 4 - 1); ++i) {
       nop();
     }
 
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb000081e);
@@ -178,33 +178,33 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
 
 
   {
-    // Will generate (void)adrp.
+    // Will generate adrp.
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
-    // Move (void)adrp 1MB away.
+    // Move adrp 1MB away.
     for (size_t i = 0; i < (1 * 1024 * 1024 / 4); ++i) {
       nop();
     }
 
-    (void)LongAddressGen(Reg::r30, &Label);
+    LongAddressGen(Reg::r30, &Label);
     nop();
     CHECK(DisassembleEncoding(262145) == 0x90fff81e);
     CHECK(DisassembleEncoding(262146) == 0xd503201f);
   }
 
   {
-    // Will generate nop + (void)adrp.
+    // Will generate nop + adrp.
     BiDirectionalLabel Label;
-    (void)LongAddressGen(Reg::r30, &Label);
+    LongAddressGen(Reg::r30, &Label);
 
     // Move label 1MB away, plus a page, and then aligned to a page.
     for (size_t i = 0; i < ((1 * 1024 * 1024 + 4096) / 4 - 2); ++i) {
       nop();
     }
 
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd503201f);
@@ -212,16 +212,16 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   }
 
   {
-    // Will generate (void)adrp + add.
+    // Will generate adrp + add.
     BiDirectionalLabel Label;
-    (void)LongAddressGen(Reg::r30, &Label);
+    LongAddressGen(Reg::r30, &Label);
 
     // Move label 1MB away, plus a page, plus one instruction.
     for (size_t i = 0; i < ((1 * 1024 * 1024 + 4096) / 4 - 1); ++i) {
       nop();
     }
 
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb000081e);

FEXCore/unittests/Emitter/Branch_Tests.cpp

@@ -9,17 +9,17 @@ using namespace ARMEmitter;
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Conditional branch immediate") {
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)b(Condition::CC_PL, &Label);
+    b(Condition::CC_PL, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x54ffffe5);
   }
 
   {
     ForwardLabel Label;
-    (void)b(Condition::CC_PL, &Label);
-    (void)Bind(&Label);
+    b(Condition::CC_PL, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x54000025);
@@ -27,17 +27,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Conditional branch immediat
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)b(Condition::CC_PL, &Label);
+    b(Condition::CC_PL, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x54ffffe5);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)b(Condition::CC_PL, &Label);
-    (void)Bind(&Label);
+    b(Condition::CC_PL, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x54000025);
@@ -46,17 +46,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Conditional branch immediat
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Branch consistent conditional") {
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)bc(Condition::CC_PL, &Label);
+    bc(Condition::CC_PL, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x54fffff5);
   }
 
   {
     ForwardLabel Label;
-    (void)bc(Condition::CC_PL, &Label);
-    (void)Bind(&Label);
+    bc(Condition::CC_PL, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x54000035);
@@ -64,17 +64,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Branch consistent condition
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)bc(Condition::CC_PL, &Label);
+    bc(Condition::CC_PL, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x54fffff5);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)bc(Condition::CC_PL, &Label);
-    (void)Bind(&Label);
+    bc(Condition::CC_PL, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x54000035);
@@ -89,17 +89,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch regist
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immediate") {
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)b(&Label);
+    b(&Label);
 
     CHECK(DisassembleEncoding(1) == 0x17ffffff);
   }
 
   {
     ForwardLabel Label;
-    (void)b(&Label);
-    (void)Bind(&Label);
+    b(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x14000001);
@@ -107,17 +107,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immedi
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)b(&Label);
+    b(&Label);
 
     CHECK(DisassembleEncoding(1) == 0x17ffffff);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)b(&Label);
-    (void)Bind(&Label);
+    b(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x14000001);
@@ -125,17 +125,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immedi
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)bl(&Label);
+    bl(&Label);
 
     CHECK(DisassembleEncoding(1) == 0x97ffffff);
   }
 
   {
     ForwardLabel Label;
-    (void)bl(&Label);
-    (void)Bind(&Label);
+    bl(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x94000001);
@@ -143,17 +143,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immedi
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)bl(&Label);
+    bl(&Label);
 
     CHECK(DisassembleEncoding(1) == 0x97ffffff);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)bl(&Label);
-    (void)Bind(&Label);
+    bl(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x94000001);
@@ -162,17 +162,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immedi
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)cbz(Size::i32Bit, Reg::r29, &Label);
+    cbz(Size::i32Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x34fffffd);
   }
 
   {
     ForwardLabel Label;
-    (void)cbz(Size::i32Bit, Reg::r29, &Label);
-    (void)Bind(&Label);
+    cbz(Size::i32Bit, Reg::r29, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3400003d);
@@ -180,17 +180,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)cbz(Size::i32Bit, Reg::r29, &Label);
+    cbz(Size::i32Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x34fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)cbz(Size::i32Bit, Reg::r29, &Label);
-    (void)Bind(&Label);
+    cbz(Size::i32Bit, Reg::r29, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3400003d);
@@ -198,17 +198,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)cbz(Size::i64Bit, Reg::r29, &Label);
+    cbz(Size::i64Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb4fffffd);
   }
 
   {
     ForwardLabel Label;
-    (void)cbz(Size::i64Bit, Reg::r29, &Label);
-    (void)Bind(&Label);
+    cbz(Size::i64Bit, Reg::r29, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb400003d);
@@ -216,17 +216,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)cbz(Size::i64Bit, Reg::r29, &Label);
+    cbz(Size::i64Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb4fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)cbz(Size::i64Bit, Reg::r29, &Label);
-    (void)Bind(&Label);
+    cbz(Size::i64Bit, Reg::r29, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb400003d);
@@ -234,17 +234,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)cbnz(Size::i32Bit, Reg::r29, &Label);
+    cbnz(Size::i32Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x35fffffd);
   }
 
   {
     ForwardLabel Label;
-    (void)cbnz(Size::i32Bit, Reg::r29, &Label);
-    (void)Bind(&Label);
+    cbnz(Size::i32Bit, Reg::r29, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3500003d);
@@ -252,17 +252,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)cbnz(Size::i32Bit, Reg::r29, &Label);
+    cbnz(Size::i32Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x35fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)cbnz(Size::i32Bit, Reg::r29, &Label);
-    (void)Bind(&Label);
+    cbnz(Size::i32Bit, Reg::r29, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3500003d);
@@ -270,17 +270,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)cbnz(Size::i64Bit, Reg::r29, &Label);
+    cbnz(Size::i64Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb5fffffd);
   }
 
   {
     ForwardLabel Label;
-    (void)cbnz(Size::i64Bit, Reg::r29, &Label);
-    (void)Bind(&Label);
+    cbnz(Size::i64Bit, Reg::r29, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb500003d);
@@ -288,17 +288,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)cbnz(Size::i64Bit, Reg::r29, &Label);
+    cbnz(Size::i64Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb5fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)cbnz(Size::i64Bit, Reg::r29, &Label);
-    (void)Bind(&Label);
+    cbnz(Size::i64Bit, Reg::r29, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb500003d);
@@ -307,17 +307,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate") {
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)tbz(Reg::r29, 0, &Label);
+    tbz(Reg::r29, 0, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x3607fffd);
   }
 
   {
     ForwardLabel Label;
-    (void)tbz(Reg::r29, 0, &Label);
-    (void)Bind(&Label);
+    tbz(Reg::r29, 0, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3600003d);
@@ -325,17 +325,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)tbz(Reg::r29, 0, &Label);
+    tbz(Reg::r29, 0, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x3607fffd);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)tbz(Reg::r29, 0, &Label);
-    (void)Bind(&Label);
+    tbz(Reg::r29, 0, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3600003d);
@@ -343,17 +343,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)tbz(Reg::r29, 63, &Label);
+    tbz(Reg::r29, 63, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb6fffffd);
   }
 
   {
     ForwardLabel Label;
-    (void)tbz(Reg::r29, 63, &Label);
-    (void)Bind(&Label);
+    tbz(Reg::r29, 63, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb6f8003d);
@@ -361,17 +361,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)tbz(Reg::r29, 63, &Label);
+    tbz(Reg::r29, 63, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb6fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)tbz(Reg::r29, 63, &Label);
-    (void)Bind(&Label);
+    tbz(Reg::r29, 63, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb6f8003d);
@@ -379,17 +379,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)tbnz(Reg::r29, 0, &Label);
+    tbnz(Reg::r29, 0, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x3707fffd);
   }
 
   {
     ForwardLabel Label;
-    (void)tbnz(Reg::r29, 0, &Label);
-    (void)Bind(&Label);
+    tbnz(Reg::r29, 0, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3700003d);
@@ -397,17 +397,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)tbnz(Reg::r29, 0, &Label);
+    tbnz(Reg::r29, 0, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x3707fffd);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)tbnz(Reg::r29, 0, &Label);
-    (void)Bind(&Label);
+    tbnz(Reg::r29, 0, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3700003d);
@@ -415,17 +415,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)tbnz(Reg::r29, 63, &Label);
+    tbnz(Reg::r29, 63, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb7fffffd);
   }
 
   {
     ForwardLabel Label;
-    (void)tbnz(Reg::r29, 63, &Label);
-    (void)Bind(&Label);
+    tbnz(Reg::r29, 63, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb7f8003d);
@@ -433,17 +433,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BiDirectionalLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
-    (void)tbnz(Reg::r29, 63, &Label);
+    tbnz(Reg::r29, 63, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb7fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    (void)tbnz(Reg::r29, 63, &Label);
-    (void)Bind(&Label);
+    tbnz(Reg::r29, 63, &Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb7f8003d);

FEXCore/unittests/Emitter/Loadstore_Tests.cpp

@@ -1323,7 +1323,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: LDAPR/STLR unscaled imme
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal") {
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
     ldr(WReg::w30, &Label);
 
@@ -1332,7 +1332,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
     ldr(SReg::s30, &Label);
 
@@ -1341,7 +1341,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
     ldr(XReg::x30, &Label);
 
@@ -1350,7 +1350,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
     ldr(DReg::d30, &Label);
 
@@ -1359,7 +1359,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
     ldrsw(XReg::x30, &Label);
 
@@ -1368,7 +1368,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
     ldr(QReg::q30, &Label);
 
@@ -1377,7 +1377,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
     prfm(Prefetch::PLDL1KEEP, &Label);
 
@@ -1387,7 +1387,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(WReg::w30, &Label);
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x1800003e);
@@ -1396,7 +1396,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(SReg::s30, &Label);
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x1c00003e);
@@ -1405,7 +1405,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(XReg::x30, &Label);
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x5800003e);
@@ -1414,7 +1414,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(DReg::d30, &Label);
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x5c00003e);
@@ -1423,7 +1423,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldrsw(XReg::x30, &Label);
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x9800003e);
@@ -1432,7 +1432,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(QReg::q30, &Label);
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x9c00003e);
@@ -1441,7 +1441,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     prfm(Prefetch::PLDL1KEEP, &Label);
-    (void)Bind(&Label);
+    Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd8000020);

FEXHeaderUtils/FEXHeaderUtils/BitUtils.h

@@ -7,7 +7,6 @@
 #include <climits>
 #include <cstddef>
 #include <cstdint>
-#include <cstring>
 #include <type_traits>
 
 namespace FEXCore {
@@ -66,17 +65,4 @@ constexpr int FindFirstSetBit(T value) noexcept {
   return trailing_zeroes + 1;
 }
 
-// Stand-in for std::bit_cast until libc++ implements it.
-template<typename To, typename From>
-[[nodiscard]]
-inline To BitCast(const From& source) noexcept {
-  static_assert(sizeof(From) == sizeof(To), "BitCast source and destination types must be equal in size.");
-  static_assert(std::is_trivially_copyable_v<From>, "BitCast source type must be trivially copyable.");
-  static_assert(std::is_trivially_copyable_v<To>, "BitCast destination type must be trivially copyable.");
-
-  std::aligned_storage_t<sizeof(To), alignof(To)> storage;
-  std::memcpy(&storage, &source, sizeof(storage));
-  return reinterpret_cast<To&>(storage);
-}
-
 } // namespace FEXCore

Source/Common/HostFeatures.cpp

@@ -436,6 +436,7 @@ static void OverrideFeatures(FEXCore::HostFeatures* Features, uint64_t ForceSVEW
   ENABLE_DISABLE_OPTION(SupportsPreserveAllABI, PRESERVEALLABI, PRESERVEALLABI);
   ENABLE_DISABLE_OPTION(SupportsWFXT, WFXT, WFXT);
   ENABLE_DISABLE_OPTION(Supports3DNow, 3DNOW, 3DNOW);
+  ENABLE_DISABLE_OPTION(SupportsSSE4a, SSE4A, SSE4A);
   GET_SINGLE_OPTION(Crypto, CRYPTO);
 
 #undef ENABLE_DISABLE_OPTION

Source/Common/SHMStats.h

@@ -26,6 +26,9 @@ static inline void store_memory_barrier() {
 
 namespace FEX::SHMStats {
 class StatAllocBase {
+public:
+  virtual ~StatAllocBase() = default;
+
 protected:
   FEXCore::SHMStats::ThreadStats* AllocateSlot(uint32_t TID);
   void DeallocateSlot(FEXCore::SHMStats::ThreadStats* AllocatedSlot);

Source/Common/VolatileMetadata.h

@@ -16,4 +16,36 @@ struct ExtendedVolatileMetadata {
 };
 
 fextl::unordered_map<fextl::string, ExtendedVolatileMetadata> ParseExtendedVolatileMetadata(std::string_view ListOfDescriptors);
+
+inline void ApplyFEXExtendedVolatileMetadata(FEX::VolatileMetadata::ExtendedVolatileMetadata& ExtendedMetaData,
+                                             fextl::set<uint64_t>& VolatileInstructions, FEXCore::IntervalList<uint64_t>& VolatileValidRanges,
+                                             uint64_t Address, uint64_t EndAddress, uint64_t FileOffset = 0, uint64_t FileOffsetEnd = ~0ULL) {
+  // Load FEX extended volatile metadata.
+  // Walk the volatile instructions first if they exist.
+  for (const auto it_inst : ExtendedMetaData.VolatileInstructions) {
+    const auto inst_address = it_inst + Address;
+    if (inst_address < EndAddress) {
+      VolatileInstructions.emplace(Address + it_inst);
+    } else {
+      LogMan::Msg::DFmt("Volatile instruction 0x{:x} couldn't fit in to module range [0x{:x}, 0x{:x}). Not adding anymore volatile "
+                        "instructions. Inspect your config!",
+                        inst_address, Address, EndAddress);
+      return;
+    }
+  }
+
+  // Walk the volatile list
+  for (const auto it_ranges : ExtendedMetaData.VolatileValidRanges) {
+    if (it_ranges.Offset >= FileOffset && it_ranges.End < FileOffsetEnd) {
+      VolatileValidRanges.Insert({Address + it_ranges.Offset - FileOffset, Address + it_ranges.End - FileOffset});
+    }
+  }
+
+  // If it is fully disabled, then set the entire module range
+  if (ExtendedMetaData.ModuleTSODisabled) {
+    VolatileValidRanges.Clear();
+    VolatileValidRanges.Insert({Address, EndAddress});
+  }
+}
+
 } // namespace FEX::VolatileMetadata

Source/Tools/FEXConfig/main.qml

@@ -721,10 +721,17 @@ ApplicationWindow {
             }
 
             ConfigCheckBox {
+                id: x87ReducedPrecisionCheckbox
                 text: qsTr("Reduced x87 precision")
                 config: "X87ReducedPrecision"
             }
 
+            ConfigCheckBox {
+                text: qsTr("Strict reduced x87 precision")
+                config: "X87StrictReducedPrecision"
+                enabled: x87ReducedPrecisionCheckbox.checked
+            }
+
             ConfigCheckBox {
                 text: qsTr("Unsafe local flags optimization")
                 config: "ABILocalFlags"

Source/Tools/FEXGetConfig/Main.cpp

@@ -2,9 +2,9 @@
 #include "Common/cpp-optparse/OptionParser.h"
 #include "Common/Config.h"
 #include "Common/FEXServerClient.h"
-#include "FEXHeaderUtils/Filesystem.h"
 #include "git_version.h"
 #include <FEXCore/Config/Config.h>
+#include <FEXCore/Utils/PrctlUtils.h>
 #include <FEXCore/fextl/fmt.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXHeaderUtils/Filesystem.h>
@@ -23,19 +23,6 @@ struct TSOEmulationFacts {
 
 #ifdef _M_ARM_64
 bool CheckForHardwareTSO() {
-  // We need to check if these are defined or not. This is a very fresh feature.
-#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
   // Check to see if this is supported.
   auto Result = prctl(PR_GET_MEM_MODEL, 0, 0, 0, 0);
   if (Result == -1) {

Source/Tools/FEXInterpreter/FEXInterpreter.cpp

@@ -31,6 +31,7 @@ $end_info$
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Telemetry.h>
 #include <FEXCore/Utils/Threads.h>
+#include <FEXCore/Utils/PrctlUtils.h>
 #include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/fextl/fmt.h>
 #include <FEXCore/fextl/memory.h>
@@ -160,19 +161,6 @@ bool QueryInterpreterInstalled(bool ExecutedWithFD, const FEX::Config::PortableI
 
 namespace FEX::TSO {
 void SetupTSOEmulation(FEXCore::Context::Context* CTX) {
-  // We need to check if these are defined or not. This is a very fresh feature.
-#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
   // Check to see if this is supported.
   auto Result = prctl(PR_GET_MEM_MODEL, 0, 0, 0, 0);
   if (Result == -1) {
@@ -201,19 +189,6 @@ void SetupTSOEmulation(FEXCore::Context::Context* CTX) {
 
 namespace FEX::CompatInput {
 void SetupCompatInput(bool enable) {
-  // We need to check if these are defined or not. This is a very fresh feature.
-#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
   // Check to see if this is supported.
   auto Result = prctl(PR_GET_COMPAT_INPUT, 0, 0, 0, 0);
   if (Result == -1) {
@@ -231,15 +206,6 @@ void SetupCompatInput(bool enable) {
 
 namespace FEX::GCS {
 void CheckForGCS() {
-#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
   uint64_t ShadowStackWord {};
   if (prctl(PR_GET_SHADOW_STACK_STATUS, &ShadowStackWord, 0, 0, 0) == -1) {
     return;
@@ -256,28 +222,6 @@ void CheckForGCS() {
 }
 } // namespace FEX::GCS
 
-namespace FEX::UnalignedAtomic {
-void SetupKernelUnalignedAtomics() {
-#ifndef PR_ARM64_SET_UNALIGN_ATOMIC
-#define PR_ARM64_SET_UNALIGN_ATOMIC 0x46455849
-#define PR_ARM64_UNALIGN_ATOMIC_EMULATE (1UL << 0)
-#define PR_ARM64_UNALIGN_ATOMIC_BACKPATCH (1UL << 1)
-#define PR_ARM64_UNALIGN_ATOMIC_STRICT_SPLIT_LOCKS (1UL << 2)
-#endif
-
-  // Interfaces with downstream FEX kernel patches to control unaligned atomic handling
-  FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
-  FEX_CONFIG_OPT(StrictInProcessSplitLocks, STRICTINPROCESSSPLITLOCKS);
-
-  uint64_t Flags = (StrictInProcessSplitLocks() ? PR_ARM64_UNALIGN_ATOMIC_STRICT_SPLIT_LOCKS : 0) |
-                   (ParanoidTSO() ? 0 : PR_ARM64_UNALIGN_ATOMIC_BACKPATCH) | PR_ARM64_UNALIGN_ATOMIC_EMULATE;
-
-  if (prctl(PR_ARM64_SET_UNALIGN_ATOMIC, Flags, 0, 0, 0) != -1) {
-    LogMan::Msg::IFmt("FEX: Kernel unaligned atomics enabled!");
-  }
-}
-} // namespace FEX::UnalignedAtomic
-
 /**
  * @brief Get an FD from an environment variable and then unset the environment variable.
  *
@@ -502,12 +446,7 @@ int main(int argc, char** argv, char** const envp) {
     free(data);
   }
 
-  {
-    FEX_CONFIG_OPT(TraceProfiler, TRACEPROFILER);
-    if (TraceProfiler()) {
-      FEXCore::Profiler::Init(Program.ProgramName, Program.ProgramPath);
-    }
-  }
+  FEXCore::Profiler::Init(Program.ProgramName, Program.ProgramPath);
 
   bool SupportsAVX {};
   fextl::unique_ptr<FEXCore::Context::Context> CTX;
@@ -519,7 +458,6 @@ int main(int argc, char** argv, char** const envp) {
 
   // Setup TSO hardware emulation immediately after initializing the context.
   FEX::TSO::SetupTSOEmulation(CTX.get());
-  FEX::UnalignedAtomic::SetupKernelUnalignedAtomics();
 
   if (!Loader.Is64BitMode()) {
     // Tell the kernel we want to use the compat input syscalls even though we're

Source/Tools/LinuxEmulation/LinuxSyscalls/Seccomp/BPFEmitter.cpp

@@ -164,7 +164,7 @@ uint64_t BPFEmitter::HandleJmp(uint32_t BPFIP, uint32_t NumInst, const sock_filt
       TargetLabel = JumpLabels.try_emplace(Target, ARMEmitter::ForwardLabel {}).first;
     }
 
-    EMIT_INST((void)b(&TargetLabel->second));
+    EMIT_INST(b(&TargetLabel->second));
     break;
   }
   case BPF_JEQ:
@@ -208,8 +208,8 @@ uint64_t BPFEmitter::HandleJmp(uint32_t BPFIP, uint32_t NumInst, const sock_filt
       TargetFalseLabel = JumpLabels.try_emplace(TargetFalse, ARMEmitter::ForwardLabel {}).first;
     }
 
-    EMIT_INST((void)b(CompareResultOp, &TargetTrueLabel->second));
-    EMIT_INST((void)b(&TargetFalseLabel->second));
+    EMIT_INST(b(CompareResultOp, &TargetTrueLabel->second));
+    EMIT_INST(b(&TargetFalseLabel->second));
     break;
   }
   default: RETURN_ERROR(-EINVAL); // Unknown jump type
@@ -263,7 +263,7 @@ uint64_t BPFEmitter::HandleEmission(uint32_t flags, const sock_fprog* prog) {
     if constexpr (!CalculateSize) {
       auto jump_label = JumpLabels.find(i);
       if (jump_label != JumpLabels.end()) {
-        (void)Bind(&jump_label->second);
+        Bind(&jump_label->second);
       }
     }
 
@@ -359,7 +359,7 @@ uint64_t BPFEmitter::JITFilter(uint32_t flags, const sock_fprog* prog) {
   // Emit the constant pool.
   Align();
   for (auto& Const : ConstPool) {
-    (void)Bind(&Const.second);
+    Bind(&Const.second);
     dc32(Const.first);
   }
 

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls.cpp

@@ -771,6 +771,9 @@ SyscallHandler::SyscallHandler(FEXCore::Context::Context* _CTX, FEX::HLE::Signal
   Alloc32Handler = FEX::HLE::Create32BitAllocator();
 
   SignalDelegation->RegisterHostSignalHandler(SIGSEGV, HandleSegfault, true);
+
+  FEX_CONFIG_OPT(ExtendedVolatileMetadataConfig, EXTENDEDVOLATILEMETADATA);
+  ExtendedMetaData = FEX::VolatileMetadata::ParseExtendedVolatileMetadata(ExtendedVolatileMetadataConfig());
 }
 
 SyscallHandler::~SyscallHandler() {
@@ -905,9 +908,7 @@ static bool isHEX(char c) {
   return (c >= '0' && c <= '9') || (c >= 'a' && c <= 'f');
 }
 
-fextl::unique_ptr<FEXCore::HLE::SourcecodeMap>
-SyscallHandler::GenerateMap(const std::string_view& GuestBinaryFile, const std::string_view& GuestBinaryFileId) {
-
+fextl::unique_ptr<FEXCore::HLE::SourcecodeMap> SyscallHandler::GenerateMap(std::string_view GuestBinaryFile, std::string_view GuestBinaryFileId) {
   ELFParser GuestELF;
 
   if (!GuestELF.ReadElf(fextl::string(GuestBinaryFile))) {

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls.h

@@ -8,6 +8,7 @@ $end_info$
 
 #pragma once
 
+#include "Common/VolatileMetadata.h"
 #include "LinuxSyscalls/FileManagement.h"
 #include "LinuxSyscalls/LinuxAllocator.h"
 #include "LinuxSyscalls/ThreadManager.h"
@@ -78,7 +79,6 @@ void RegisterFS(FEX::HLE::SyscallHandler* Handler);
 void RegisterInfo(FEX::HLE::SyscallHandler* Handler);
 void RegisterIO(FEX::HLE::SyscallHandler* Handler);
 void RegisterMemory(FEX::HLE::SyscallHandler* Handler);
-void RegisterNuma(FEX::HLE::SyscallHandler* Handler);
 void RegisterSignals(FEX::HLE::SyscallHandler* Handler);
 void RegisterThread(FEX::HLE::SyscallHandler* Handler);
 void RegisterTimer(FEX::HLE::SyscallHandler* Handler);
@@ -128,8 +128,6 @@ public:
   using SyscallPtrArg6 = uint64_t (*)(FEXCore::Core::CpuStateFrame* Frame, uint64_t, uint64_t, uint64_t, uint64_t, uint64_t, uint64_t);
 
   struct SyscallFunctionDefinition {
-    uint8_t NumArgs;
-    FEXCore::IR::SyscallFlags Flags;
     union {
       void* Ptr;
       SyscallPtrArg0 Ptr0;
@@ -141,6 +139,8 @@ public:
       SyscallPtrArg6 Ptr6;
     };
     int32_t HostSyscallNumber;
+    uint8_t NumArgs;
+    FEXCore::IR::SyscallFlags Flags;
 #ifdef DEBUG_STRACE
     fextl::string StraceFmt;
 #endif
@@ -253,7 +253,12 @@ public:
   uint64_t GuestShmdt(bool Is64Bit, FEXCore::Core::InternalThreadState*, const void* shmaddr);
 
   ///// Memory Manager tracking /////
-  void TrackMmap(FEXCore::Core::InternalThreadState* Thread, uint64_t addr, size_t length, int prot, int flags, int fd, off_t offset);
+  struct LateApplyExtendedVolatileMetadata {
+    fextl::set<uint64_t> VolatileInstructions {};
+    FEXCore::IntervalList<uint64_t> VolatileValidRanges {};
+  };
+  std::optional<LateApplyExtendedVolatileMetadata>
+  TrackMmap(FEXCore::Core::InternalThreadState* Thread, uint64_t addr, size_t length, int prot, int flags, int fd, off_t offset);
   void TrackMunmap(FEXCore::Core::InternalThreadState* Thread, void* addr, size_t length);
   void TrackMremap(FEXCore::Core::InternalThreadState* Thread, uint64_t OldAddress, size_t OldSize, size_t NewSize, int flags, uint64_t NewAddress);
   void TrackShmat(FEXCore::Core::InternalThreadState* Thread, int shmid, uint64_t shmaddr, int shmflg, uint64_t Length);
@@ -330,8 +335,8 @@ protected:
 
   fextl::vector<SyscallFunctionDefinition> Definitions {std::max<std::size_t>(FEX::HLE::x64::SYSCALL_x64_MAX, FEX::HLE::x32::SYSCALL_x86_MAX),
                                                         {
-                                                          .NumArgs = 255,
                                                           .Ptr = reinterpret_cast<void*>(&UnimplementedSyscall),
+                                                          .NumArgs = 255,
                                                         }};
   std::mutex MMapMutex;
 
@@ -345,11 +350,13 @@ protected:
   uint32_t GuestKernelVersion {};
 
   FEXCore::Context::Context* CTX;
-private:
 
+private:
   FEX::HLE::SignalDelegator* SignalDelegation;
   FEX::HLE::ThunkHandler* ThunkHandler;
 
+  fextl::unordered_map<fextl::string, FEX::VolatileMetadata::ExtendedVolatileMetadata> ExtendedMetaData {};
+
   std::mutex FutexMutex;
   std::mutex SyscallMutex;
   FEX::CodeLoader* LocalLoader {};
@@ -358,17 +365,12 @@ private:
 #ifdef DEBUG_STRACE
   void Strace(FEXCore::HLE::SyscallArguments* Args, uint64_t Ret);
 #endif
+  fextl::unique_ptr<FEXCore::HLE::SourcecodeMap> GenerateMap(std::string_view GuestBinaryFile, std::string_view GuestBinaryFileId) override;
 
   fextl::unique_ptr<FEX::HLE::MemAllocator> Alloc32Handler {};
-
-  fextl::unique_ptr<FEXCore::HLE::SourcecodeMap>
-  GenerateMap(const std::string_view& GuestBinaryFile, const std::string_view& GuestBinaryFileId) override;
-
   std::atomic<uint64_t> AnonSharedId {1};
 };
 
-uint64_t HandleSyscall(SyscallHandler* Handler, FEXCore::Core::CpuStateFrame* Frame, FEXCore::HLE::SyscallArguments* Args);
-
 #define SYSCALL_ERRNO()              \
   do {                               \
     if (Result == -1) return -errno; \

Source/Tools/LinuxEmulation/LinuxSyscalls/SyscallsSMCTracking.cpp

@@ -22,6 +22,7 @@ $end_info$
 #include <FEXCore/Utils/MathUtils.h>
 #include <FEXCore/Utils/SignalScopeGuards.h>
 #include <FEXCore/Utils/TypeDefines.h>
+#include <FEXHeaderUtils/Filesystem.h>
 
 namespace FEX::HLE {
 // SMC interactions
@@ -200,6 +201,7 @@ void* SyscallHandler::GuestMmap(bool Is64Bit, FEXCore::Core::InternalThreadState
 
   uint64_t Result {};
   size_t Size = FEXCore::AlignUp(length, FEXCore::Utils::FEX_PAGE_SIZE);
+  std::optional<LateApplyExtendedVolatileMetadata> LateMetadata = std::nullopt;
 
   {
     // NOTE: Frontend calls this with a nullptr Thread during initialization, but
@@ -221,10 +223,16 @@ void* SyscallHandler::GuestMmap(bool Is64Bit, FEXCore::Core::InternalThreadState
       }
     }
 
-    FEX::HLE::_SyscallHandler->TrackMmap(Thread, Result, length, prot, flags, fd, offset);
+    LateMetadata = FEX::HLE::_SyscallHandler->TrackMmap(Thread, Result, length, prot, flags, fd, offset);
   }
 
   FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessary(Thread, Result, Size);
+
+  if (LateMetadata) {
+    auto CodeInvalidationlk = GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), Thread);
+    CTX->AddForceTSOInformation(LateMetadata->VolatileValidRanges, std::move(LateMetadata->VolatileInstructions));
+  }
+
   return reinterpret_cast<void*>(Result);
 }
 
@@ -256,6 +264,11 @@ uint64_t SyscallHandler::GuestMunmap(bool Is64Bit, FEXCore::Core::InternalThread
   }
   FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessary(Thread, reinterpret_cast<uint64_t>(addr), Size);
 
+  if (length) {
+    auto CodeInvalidationlk = GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), Thread);
+    CTX->RemoveForceTSOInformation(reinterpret_cast<uint64_t>(addr), length);
+  }
+
   return Result;
 }
 
@@ -360,11 +373,15 @@ uint64_t SyscallHandler::GuestShmdt(bool Is64Bit, FEXCore::Core::InternalThreadS
 }
 
 // MMan Tracking
-void SyscallHandler::TrackMmap(FEXCore::Core::InternalThreadState* Thread, uint64_t addr, size_t length, int prot, int flags, int fd, off_t offset) {
+std::optional<SyscallHandler::LateApplyExtendedVolatileMetadata> SyscallHandler::TrackMmap(
+  FEXCore::Core::InternalThreadState* Thread, uint64_t addr, size_t length, int prot, int flags, int fd, off_t offset) {
   size_t Size = FEXCore::AlignUp(length, FEXCore::Utils::FEX_PAGE_SIZE);
+  const auto ProtMapping = VMATracking::VMAProt::fromProt(prot);
 
   VMATracking::MappedResource* Resource = nullptr;
 
+  std::optional<SyscallHandler::LateApplyExtendedVolatileMetadata> VolatileMetadata = std::nullopt;
+
   if (!(flags & MAP_ANONYMOUS)) {
     struct stat64 buf;
     fstat64(fd, &buf);
@@ -382,6 +399,21 @@ void SyscallHandler::TrackMmap(FEXCore::Core::InternalThreadState* Thread, uint6
         Resource->MappedFile->Filename = fextl::string(Tmp, PathLength);
         Resource->Iterator = Iter;
       }
+
+      const fextl::string Filename = FHU::Filesystem::GetFilename(Resource->MappedFile->Filename);
+
+      // We now have the filename and the offset in the filename getting mapped.
+      // Check for extended volatile metadata.
+      auto it = ExtendedMetaData.find(Filename);
+      if (it != ExtendedMetaData.end()) {
+        SyscallHandler::LateApplyExtendedVolatileMetadata LateMetadata;
+        FEX::VolatileMetadata::ApplyFEXExtendedVolatileMetadata(
+          it->second, LateMetadata.VolatileInstructions, LateMetadata.VolatileValidRanges, addr, addr + length, offset, offset + length);
+
+        if (!LateMetadata.VolatileInstructions.empty() || !LateMetadata.VolatileValidRanges.Empty()) {
+          VolatileMetadata.emplace(std::move(LateMetadata));
+        }
+      }
     }
   } else if (flags & MAP_SHARED) {
     VMATracking::MRID mrid {VMATracking::SpecialDev::Anon, AnonSharedId++};
@@ -394,7 +426,8 @@ void SyscallHandler::TrackMmap(FEXCore::Core::InternalThreadState* Thread, uint6
     Resource = nullptr;
   }
 
-  VMATracking.TrackVMARange(CTX, Resource, addr, offset, Size, VMATracking::VMAFlags::fromFlags(flags), VMATracking::VMAProt::fromProt(prot));
+  VMATracking.TrackVMARange(CTX, Resource, addr, offset, Size, VMATracking::VMAFlags::fromFlags(flags), ProtMapping);
+  return VolatileMetadata;
 }
 
 void SyscallHandler::TrackMunmap(FEXCore::Core::InternalThreadState* Thread, void* addr, size_t length) {

Source/Tools/LinuxEmulation/LinuxSyscalls/ThreadManager.cpp

@@ -46,7 +46,7 @@ void ThreadManager::StatAlloc::Initialize() {
   // 1 page: 99 slots
   // 1 MB: 26211 slots
   // 128 MB: 3355440 slots
-  Base = ::mmap(nullptr, MAX_STATS_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0);
+  Base = FEXCore::Allocator::mmap(nullptr, MAX_STATS_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0);
   if (Base == MAP_FAILED) {
     LogMan::Msg::EFmt("[StatAlloc] mmap base failed");
     Base = nullptr;
@@ -55,10 +55,10 @@ void ThreadManager::StatAlloc::Initialize() {
 
   // Allocate a small working shared space for now, grow as necessary.
   {
-    auto SharedBase = ::mmap(Base, CurrentSize, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, fd, 0);
+    auto SharedBase = FEXCore::Allocator::mmap(Base, CurrentSize, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, fd, 0);
     if (SharedBase == MAP_FAILED) {
       LogMan::Msg::EFmt("[StatAlloc] mmap shm failed");
-      munmap(Base, MAX_STATS_SIZE);
+      FEXCore::Allocator::munmap(Base, MAX_STATS_SIZE);
       Base = nullptr;
       goto err;
     }
@@ -89,7 +89,7 @@ uint32_t ThreadManager::StatAlloc::FrontendAllocateSlots(uint32_t NewSize) {
   }
 
   {
-    auto SharedBase = ::mmap(Base, NewSize, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, fd, 0);
+    auto SharedBase = FEXCore::Allocator::mmap(Base, NewSize, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, fd, 0);
     if (SharedBase == MAP_FAILED) {
       LogMan::Msg::EFmt("[StatAlloc] allocate more mmap shm failed");
       goto err;
@@ -140,7 +140,7 @@ void ThreadManager::StatAlloc::UnlockAfterFork(FEXCore::Core::InternalThreadStat
 
   // shm_memory ownership is retained by the parent process, so the child must replace it with its own one.
   // Otherwise this process will keep reporting in the original parent thread's stats region.
-  munmap(Base, MAX_STATS_SIZE);
+  FEXCore::Allocator::munmap(Base, MAX_STATS_SIZE);
   Base = nullptr;
   CurrentSize = 0;
   Head = nullptr;
@@ -173,7 +173,8 @@ FEX::HLE::ThreadStateObject* ThreadManager::CreateThread(uint64_t InitialRIP, ui
   auto Frame = ThreadStateObject->Thread->CurrentFrame;
 
   // Allocate the call-ret stack with guard pages on both sides
-  auto AllocBase = reinterpret_cast<uint64_t>(::mmap(nullptr, CALLRET_STACK_ALLOC_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  auto AllocBase =
+    reinterpret_cast<uint64_t>(FEXCore::Allocator::mmap(nullptr, CALLRET_STACK_ALLOC_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
   // Set the base used for invalidation to the start past the guard pages
   ThreadStateObject->Thread->CallRetStackBase = reinterpret_cast<void*>(AllocBase + FEXCore::Utils::FEX_PAGE_SIZE);
   ::mprotect(ThreadStateObject->Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE, PROT_READ | PROT_WRITE);
@@ -273,7 +274,7 @@ void ThreadManager::HandleThreadDeletion(FEX::HLE::ThreadStateObject* Thread, bo
   }
 
   // Free the call-ret stack
-  ::munmap(reinterpret_cast<void*>(Thread->GetCallRetStackInfo().AllocationBase), CALLRET_STACK_ALLOC_SIZE);
+  FEXCore::Allocator::munmap(reinterpret_cast<void*>(Thread->GetCallRetStackInfo().AllocationBase), CALLRET_STACK_ALLOC_SIZE);
 
   // If the LDT segment exists then deallocate it.
   if (Thread->ldt_entry_count) {

Source/Tools/LinuxEmulation/LinuxSyscalls/ThreadManager.h

@@ -163,8 +163,6 @@ public:
     Stat.CleanupForExit();
   }
 
-  StatAlloc Stat;
-
   ///< Returns the ThreadStateObject from a CpuStateFrame object.
   static inline FEX::HLE::ThreadStateObject* GetStateObjectFromCPUState(FEXCore::Core::CpuStateFrame* Frame) {
     return static_cast<FEX::HLE::ThreadStateObject*>(Frame->Thread->FrontendPtr);
@@ -240,6 +238,7 @@ public:
   }
 
 private:
+  StatAlloc Stat;
   FEXCore::Context::Context* CTX;
   FEX::HLE::SignalDelegator* SignalDelegation;
 

Source/Tools/LinuxEmulation/LinuxSyscalls/Utils/Threads.cpp

@@ -4,7 +4,6 @@
 
 #include <FEXCore/Core/Context.h>
 #include <FEXCore/Utils/Allocator.h>
-#include <FEXCore/Utils/LongJump.h>
 #include <FEXCore/Utils/Threads.h>
 
 namespace FEX::LinuxEmulation::Threads {
@@ -190,6 +189,143 @@ __attribute__((naked)) void StackPivotAndCall(void* Arg, FEXCore::Threads::Threa
 }
 #endif
 namespace PThreads {
+  namespace LongJump {
+    // This is a custom long jump implementation that avoids the glibc implementation.
+    // This is required behaviour because glibc's fortification checks don't understand stack pivots.
+    // FEX requires a stack pivot to work through a long jump, so these two features are at odds with each other.
+#ifdef _M_ARM_64
+    struct JumpBuf {
+      // All the registers that are required by AAPCS64 to save.
+      // GPRs
+      // X19, X20, X21, X22,
+      // X23, X24, X25, X26,
+      // X27, X28, X29, X30,
+      //
+      // Lower 64-bits:
+      //  V8,  V9, V10, V11,
+      // V12, V13, V14, V15,
+      //
+      // SP,
+      uint64_t Registers[21];
+    };
+    FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
+      __asm volatile(R"(
+        // x0 contains the jumpbuffer
+        stp x19, x20, [x0, #( 0 * 8)];
+        stp x21, x22, [x0, #( 2 * 8)];
+        stp x23, x24, [x0, #( 4 * 8)];
+        stp x25, x26, [x0, #( 6 * 8)];
+        stp x27, x28, [x0, #( 8 * 8)];
+        stp x29, x30, [x0, #(10 * 8)];
+
+        // FPRs
+        stp d8,   d9, [x0, #(12 * 8)];
+        stp d10, d11, [x0, #(14 * 8)];
+        stp d12, d13, [x0, #(16 * 8)];
+        stp d14, d15, [x0, #(18 * 8)];
+
+        // Move SP in to a temporary to store.
+        mov x1, sp;
+        str x1,  [x0, #(19 * 8)];
+
+        // Return zero to signify this is the SetJump.
+        mov x0, #0;
+        ret;
+      )" ::
+                       : "memory");
+    }
+
+    [[noreturn]]
+    FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value) {
+      __asm volatile(R"(
+        // x0 contains the jumpbuffer
+        ldp x19, x20, [x0, #( 0 * 8)];
+        ldp x21, x22, [x0, #( 2 * 8)];
+        ldp x23, x24, [x0, #( 4 * 8)];
+        ldp x25, x26, [x0, #( 6 * 8)];
+        ldp x27, x28, [x0, #( 8 * 8)];
+        ldp x29, x30, [x0, #(10 * 8)];
+
+        // FPRs
+        ldp d8,   d9, [x0, #(12 * 8)];
+        ldp d10, d11, [x0, #(14 * 8)];
+        ldp d12, d13, [x0, #(16 * 8)];
+        ldp d14, d15, [x0, #(18 * 8)];
+
+        // Load SP in to temporary then move
+        ldr x0,  [x0, #(19 * 8)];
+        mov sp, x0;
+
+        // Move value in to result register
+        mov x0, x1;
+        ret;
+      )" ::
+                       : "memory");
+    }
+#else
+    struct JumpBuf {
+      // Registers to preserve
+      // RBX, RSP, RBP, R12, R13, R14, R15,
+      // <return address>
+      uint64_t Registers[8];
+    };
+
+    __attribute__((naked)) uint64_t SetJump(JumpBuf& Buffer) {
+      __asm volatile(R"(
+      .intel_syntax noprefix;
+      // rdi contains the jumpbuffer
+      mov [rdi + (0 * 8)], rbx;
+      mov [rdi + (1 * 8)], rsp;
+      mov [rdi + (2 * 8)], rbp;
+      mov [rdi + (3 * 8)], r12;
+      mov [rdi + (4 * 8)], r13;
+      mov [rdi + (5 * 8)], r14;
+      mov [rdi + (6 * 8)], r15;
+
+      // Return address is on the stack, load it and store
+      mov rsi, [rsp];
+      mov [rdi + (7 * 8)], rsi;
+
+      // Return zero to signify this is the SetJump.
+      mov rax, 0;
+      ret;
+
+      .att_syntax prefix;
+      )" ::
+                       : "memory");
+    }
+
+    [[noreturn]]
+    __attribute__((naked)) void LongJump(JumpBuf& Buffer, uint64_t Value) {
+      __asm volatile(R"(
+      .intel_syntax noprefix;
+      // rdi contains the jumpbuffer
+      mov rbx, [rdi + (0 * 8)];
+      mov rsp, [rdi + (1 * 8)];
+      mov rbp, [rdi + (2 * 8)];
+      mov r12, [rdi + (3 * 8)];
+      mov r13, [rdi + (4 * 8)];
+      mov r14, [rdi + (5 * 8)];
+      mov r15, [rdi + (6 * 8)];
+
+      // Move value in to result register
+      mov rax, rsi;
+
+      // Pop the dead return address off the stack
+      pop rsi;
+
+      // Load the original return address from the jumpbuffer
+      mov rsi, [rdi + (7 * 8)];
+
+      // Return using a jump
+      jmp rsi;
+
+      .att_syntax prefix;
+      )" ::
+                       : "memory");
+    }
+#endif
+  }; // namespace LongJump
   void* InitializeThread(void* Ptr);
 
   class PThread final : public FEXCore::Threads::Thread {
@@ -260,7 +396,7 @@ namespace PThreads {
       return STracker;
     }
 
-    void SetupLongJump(FEXCore::LongJump::JumpBuf* exit_resolver) {
+    void SetupLongJump(LongJump::JumpBuf* exit_resolver) {
       _exit_resolver = exit_resolver;
     }
 
@@ -268,7 +404,7 @@ namespace PThreads {
     void LongJumpExit(FEX::HLE::ThreadStateObject* ThreadObject, uint32_t Status) {
       this->Status = Status;
       this->ThreadObject = ThreadObject;
-      FEXCore::LongJump::LongJump(*_exit_resolver, 1);
+      LongJump::LongJump(*_exit_resolver, 1);
       FEX_UNREACHABLE;
     }
 
@@ -287,7 +423,7 @@ namespace PThreads {
     void* UserArg;
     void* Stack {};
 
-    FEXCore::LongJump::JumpBuf* _exit_resolver {};
+    LongJump::JumpBuf* _exit_resolver {};
     FEX::HLE::ThreadStateObject* ThreadObject {};
     uint32_t Status {};
   };
@@ -298,11 +434,11 @@ namespace PThreads {
     PThread* Thread {reinterpret_cast<PThread*>(Ptr)};
     StackBase = Thread->GetPivotStack();
     STracker = Thread->GetStackTracker();
-    FEXCore::LongJump::JumpBuf exit_resolver {};
+    LongJump::JumpBuf exit_resolver {};
 
     bool LongJumpExit {};
 
-    if (FEXCore::LongJump::SetJump(exit_resolver) == 0) {
+    if (LongJump::SetJump(exit_resolver) == 0) {
       Thread->SetupLongJump(&exit_resolver);
       // Run the user function.
       // `Thread` object is dead after this function returns.

Source/Tools/LinuxEmulation/LinuxSyscalls/x64/Syscalls.cpp

@@ -63,10 +63,10 @@ void x64SyscallHandler::RegisterSyscallHandlers() {
   constexpr int SYSCALL_GAP_END = 424;
 
   const SyscallFunctionDefinition InvalidSyscall {
-    .NumArgs = 0,
-    .Flags = FEXCore::IR::SyscallFlags::DEFAULT,
     .Ptr = reinterpret_cast<void*>(&UnimplementedSyscall),
     .HostSyscallNumber = SYSCALL_DEF(MAX),
+    .NumArgs = 0,
+    .Flags = FEXCore::IR::SyscallFlags::DEFAULT,
 #ifdef DEBUG_STRACE
     .StraceFmt = "Invalid",
 #endif

Source/Windows/ARM64EC/Module.cpp

@@ -39,10 +39,10 @@ $end_info$
 #include "Common/CRT/CRT.h"
 #include "Common/PortabilityInfo.h"
 #include "Common/Handle.h"
+#include "Common/VolatileMetadata.h"
 #include "DummyHandlers.h"
 #include "BTInterface.h"
 #include "Windows/Common/SHMStats.h"
-#include "Windows/Common/VolatileMetadata.h"
 
 #include <cstdint>
 #include <cstdio>
@@ -314,7 +314,7 @@ void LoadImageVolatileMetadata(const fextl::string& ModuleName, uint64_t Address
 
   auto it = ExtendedMetaData.find(ModuleName);
   if (it != ExtendedMetaData.end()) {
-    FEX::Windows::ApplyFEXExtendedVolatileMetadata(it->second, VolatileInstructions, VolatileValidRanges, Address, EndAddress);
+    FEX::VolatileMetadata::ApplyFEXExtendedVolatileMetadata(it->second, VolatileInstructions, VolatileValidRanges, Address, EndAddress);
   }
 
   if (VolatileInstructions.empty() && VolatileValidRanges.Empty()) {
@@ -640,18 +640,14 @@ NTSTATUS ProcessInit() {
 
   FEXCore::Config::Set(FEXCore::Config::CONFIG_IS64BIT_MODE, "1");
 
-  {
-    FEX_CONFIG_OPT(TraceProfiler, TRACEPROFILER);
-    if (TraceProfiler()) {
-      FEXCore::Profiler::Init("", "");
-    }
-  }
+  FEXCore::Profiler::Init("", "");
 
   FEX_CONFIG_OPT(ExtendedVolatileMetadataConfig, EXTENDEDVOLATILEMETADATA);
   ExtendedMetaData = FEX::VolatileMetadata::ParseExtendedVolatileMetadata(ExtendedVolatileMetadataConfig());
 
   SignalDelegator = fextl::make_unique<FEX::DummyHandlers::DummySignalDelegator>();
   SyscallHandler = fextl::make_unique<Exception::ECSyscallHandler>();
+  Exception::HandlerConfig.emplace();
 
   const auto NtDll = GetModuleHandle("ntdll.dll");
   const bool IsWine = !!GetProcAddress(NtDll, "wine_get_version");
@@ -665,7 +661,7 @@ NTSTATUS ProcessInit() {
   CTX->SetSignalDelegator(SignalDelegator.get());
   CTX->SetSyscallHandler(SyscallHandler.get());
   CTX->InitCore();
-  Exception::HandlerConfig.emplace(*CTX);
+
   InvalidationTracker.emplace(*CTX, Threads);
 
   HandleImageMap(NtDllBase);

Source/Windows/Common/CRT/String.cpp

@@ -56,10 +56,6 @@ long long atoll(const char*) {
   UNIMPLEMENTED();
 }
 
-long long strtoll(const char* __restrict__, char** __restrict, int) {
-  UNIMPLEMENTED();
-}
-
 long double strtold(const char* __restrict__, char** __restrict__) {
   UNIMPLEMENTED();
 }

Source/Windows/Common/TSOHandlerConfig.h

@@ -1,14 +1,13 @@
 // SPDX-License-Identifier: MIT
 #pragma once
 
-#include <FEXCore/Core/Context.h>
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Utils/ArchHelpers/Arm64.h>
 
 namespace FEX::Windows {
 class TSOHandlerConfig final {
 public:
-  TSOHandlerConfig(FEXCore::Context::Context& CTX) {
+  TSOHandlerConfig() {
     if (ParanoidTSO()) {
       UnalignedHandlerType = FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::Paranoid;
     } else if (HalfBarrierTSOEnabled()) {
@@ -16,21 +15,6 @@ public:
     } else {
       UnalignedHandlerType = FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::NonAtomic;
     }
-
-    if (TSOEnabled()) {
-      BOOL Enable = TRUE;
-      NTSTATUS Status = NtSetInformationProcess(NtCurrentProcess(), ProcessFexHardwareTso, &Enable, sizeof(Enable));
-      if (Status == STATUS_SUCCESS) {
-        CTX.SetHardwareTSOSupport(true);
-      }
-    }
-
-    uint64_t Flags = (StrictInProcessSplitLocks() ? FEX_UNALIGN_ATOMIC_STRICT_SPLIT_LOCKS : 0) |
-                     (ParanoidTSO() ? 0 : FEX_UNALIGN_ATOMIC_BACKPATCH) | FEX_UNALIGN_ATOMIC_EMULATE;
-
-    if (NtSetInformationProcess(NtCurrentProcess(), ProcessFexUnalignAtomic, &Flags, sizeof(Flags)) == STATUS_SUCCESS) {
-      LogMan::Msg::IFmt("FEX: Kernel unaligned atomics enabled!");
-    }
   }
 
   FEXCore::ArchHelpers::Arm64::UnalignedHandlerType GetUnalignedHandlerType() const {
@@ -38,10 +22,8 @@ public:
   }
 
 private:
-  FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
   FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
   FEX_CONFIG_OPT(HalfBarrierTSOEnabled, HALFBARRIERTSOENABLED);
-  FEX_CONFIG_OPT(StrictInProcessSplitLocks, STRICTINPROCESSSPLITLOCKS);
 
   FEXCore::ArchHelpers::Arm64::UnalignedHandlerType UnalignedHandlerType {FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::HalfBarrier};
 };

Source/Windows/Common/VolatileMetadata.h

@@ -1,42 +0,0 @@
-// SPDX-License-Identifier: MIT
-#pragma once
-
-#include "Common/VolatileMetadata.h"
-
-#include <FEXCore/fextl/unordered_map.h>
-#include <FEXCore/fextl/string.h>
-#include <FEXCore/fextl/set.h>
-#include <FEXCore/Utils/IntervalList.h>
-
-namespace FEX::Windows {
-inline void ApplyFEXExtendedVolatileMetadata(FEX::VolatileMetadata::ExtendedVolatileMetadata& ExtendedMetaData,
-                                             fextl::set<uint64_t>& VolatileInstructions,
-                                             FEXCore::IntervalList<uint64_t>& VolatileValidRanges, uint64_t Address, uint64_t EndAddress) {
-  // Load FEX extended volatile metadata.
-  // Walk the volatile instructions first if they exist.
-  for (const auto it_inst : ExtendedMetaData.VolatileInstructions) {
-    const auto inst_address = it_inst + Address;
-    if (inst_address < EndAddress) {
-      VolatileInstructions.emplace(Address + it_inst);
-    } else {
-      LogMan::Msg::DFmt("Volatile instruction 0x{:x} couldn't fit in to module range [0x{:x}, 0x{:x}). Not adding anymore volatile "
-                        "instructions. Inspect your config!",
-                        inst_address, Address, EndAddress);
-      return;
-    }
-  }
-
-  // Walk the volatile list
-  for (const auto it_ranges : ExtendedMetaData.VolatileValidRanges) {
-    VolatileValidRanges.Insert({Address + it_ranges.Offset, Address + it_ranges.End});
-  }
-
-  // If it is fully disabled, then set the entire module range
-  if (ExtendedMetaData.ModuleTSODisabled) {
-    VolatileValidRanges.Clear();
-    VolatileValidRanges.Insert({Address, EndAddress});
-  }
-}
-
-
-} // namespace FEX::Windows

Source/Windows/WOW64/Module.cpp

@@ -40,10 +40,10 @@ $end_info$
 #include "Common/CRT/CRT.h"
 #include "Common/PortabilityInfo.h"
 #include "Common/Handle.h"
+#include "Common/VolatileMetadata.h"
 #include "DummyHandlers.h"
 #include "BTInterface.h"
 #include "Windows/Common/SHMStats.h"
-#include "Windows/Common/VolatileMetadata.h"
 
 #include <cstdint>
 #include <type_traits>
@@ -173,7 +173,7 @@ void LoadImageVolatileMetadata(const fextl::string& ModuleName, uint64_t Address
   // Load FEX extended volatile metadata.
   auto it = ExtendedMetaData.find(ModuleName);
   if (it != ExtendedMetaData.end()) {
-    FEX::Windows::ApplyFEXExtendedVolatileMetadata(it->second, VolatileInstructions, VolatileValidRanges, Address, EndAddress);
+    FEX::VolatileMetadata::ApplyFEXExtendedVolatileMetadata(it->second, VolatileInstructions, VolatileValidRanges, Address, EndAddress);
   }
 
   if (VolatileInstructions.empty() && VolatileValidRanges.Empty()) {
@@ -516,18 +516,14 @@ void BTCpuProcessInit() {
   FEXCore::Config::Set(FEXCore::Config::CONFIG_INTERPRETER_INSTALLED, "0");
   FEXCore::Config::Set(FEXCore::Config::CONFIG_IS64BIT_MODE, "0");
 
-  {
-    FEX_CONFIG_OPT(TraceProfiler, TRACEPROFILER);
-    if (TraceProfiler()) {
-      FEXCore::Profiler::Init("", "");
-    }
-  }
+  FEXCore::Profiler::Init("", "");
 
   FEX_CONFIG_OPT(ExtendedVolatileMetadataConfig, EXTENDEDVOLATILEMETADATA);
   ExtendedMetaData = FEX::VolatileMetadata::ParseExtendedVolatileMetadata(ExtendedVolatileMetadataConfig());
 
   SignalDelegator = fextl::make_unique<FEX::DummyHandlers::DummySignalDelegator>();
   SyscallHandler = fextl::make_unique<WowSyscallHandler>();
+  Context::HandlerConfig.emplace();
   const auto NtDll = GetModuleHandle("ntdll.dll");
   const bool IsWine = !!GetProcAddress(NtDll, "wine_get_version");
   OvercommitTracker.emplace(IsWine);
@@ -540,7 +536,7 @@ void BTCpuProcessInit() {
   CTX->SetSignalDelegator(SignalDelegator.get());
   CTX->SetSyscallHandler(SyscallHandler.get());
   CTX->InitCore();
-  Context::HandlerConfig.emplace(*CTX);
+
   InvalidationTracker.emplace(*CTX, Threads);
 
   auto NtDllX86 = reinterpret_cast<SYSTEM_DLL_INIT_BLOCK*>(GetProcAddress(NtDll, "LdrSystemDllInitBlock"))->ntdll_handle;
@@ -1007,7 +1003,11 @@ NTSTATUS BTCpuNotifyMapViewOfSection(void* Unk1, void* Address, void* Unk2, SIZE
 void BTCpuNotifyUnmapViewOfSection(void* Address, BOOL After, ULONG Status) {
   if (!After) {
     ThreadCreationMutex.lock();
-    InvalidationTracker->InvalidateContainingSection(reinterpret_cast<uint64_t>(Address), true);
+    auto [Start, Size] = InvalidationTracker->InvalidateContainingSection(reinterpret_cast<uint64_t>(Address), true);
+    if (Size) {
+      std::scoped_lock Lock(CTX->GetCodeInvalidationMutex());
+      CTX->RemoveForceTSOInformation(Start, Size);
+    }
   } else {
     ThreadCreationMutex.unlock();
   }

Source/Windows/include/winternl.h

@@ -455,12 +455,6 @@ typedef enum _MEMORY_INFORMATION_CLASS {
 #define SystemEmulationBasicInformation (SYSTEM_INFORMATION_CLASS)62
 
 #define ProcessFexHardwareTso (PROCESSINFOCLASS)2000
-#define ProcessFexUnalignAtomic (PROCESSINFOCLASS)2001
-
-// These match the prctl flag values
-#define FEX_UNALIGN_ATOMIC_EMULATE (1ULL << 0)
-#define FEX_UNALIGN_ATOMIC_BACKPATCH (1ULL << 1)
-#define FEX_UNALIGN_ATOMIC_STRICT_SPLIT_LOCKS (1ULL << 2)
 
 typedef enum _KEY_VALUE_INFORMATION_CLASS {
   KeyValueBasicInformation,

docs/Readme_CN.md

@@ -26,7 +26,7 @@ FEX在ARMv8.0，ARMv8.1+和x86-64(支持AVX或更新处理器)硬件上进行过
 在AArch64 host端，用户需要准备x86-64 RootFS[创建RootFS](#RootFS-Generation)。
 
 ### 源码导览
-详见[源码大纲](docs/SourceOutline.md)。
+详见[源码大纲](SourceOutline.md)。
 
 ### 编译依赖
 * cmake (version 3.14 minimum)

docs/SourceOutline.md

@@ -1,4 +1,4 @@
-# FEX-2509
+# FEX-2510
 
 ## FEXCore
 See [FEXCore/Readme.md](../FEXCore/Readme.md) for more details
@@ -103,7 +103,7 @@ IR to IR Optimization
 - [PassManager.cpp](../FEXCore/Source/Interface/IR/PassManager.cpp): Defines which passes are run, and runs them
 - [PassManager.h](../FEXCore/Source/Interface/IR/PassManager.h)
 - [IRValidation.cpp](../FEXCore/Source/Interface/IR/Passes/IRValidation.cpp): Sanity checking pass
-- [RedundantFlagCalculationElimination.cpp](../FEXCore/Source/Interface/IR/Passes/RedundantFlagCalculationElimination.cpp): This is not used right now, possibly broken
+- [RedundantFlagCalculationElimination.cpp](../FEXCore/Source/Interface/IR/Passes/RedundantFlagCalculationElimination.cpp)
 - [RegisterAllocationPass.cpp](../FEXCore/Source/Interface/IR/Passes/RegisterAllocationPass.cpp)
 - [RegisterAllocationPass.h](../FEXCore/Source/Interface/IR/Passes/RegisterAllocationPass.h)
 

unittests/32Bit_ASM/CMakeLists.txt

@@ -38,7 +38,7 @@ foreach(ASM_SRC ${ASM_SOURCES})
 
   add_custom_command(OUTPUT ${OUTPUT_NAME}
     DEPENDS "${TMP_FILE}"
-    COMMAND "nasm" ARGS "${TMP_FILE}" "-o" "${OUTPUT_NAME}")
+    COMMAND "nasm" ARGS "-i" "${CMAKE_SOURCE_DIR}/unittests/32Bit_ASM/Includes/" "${TMP_FILE}" "-o" "${OUTPUT_NAME}")
 
   add_custom_command(OUTPUT ${OUTPUT_CONFIG_NAME}
     DEPENDS "${ASM_SRC}"

unittests/32Bit_ASM/Includes/nan_test_macros.inc

@@ -0,0 +1,130 @@
+; NaN Testing Macros for 32-bit Assembly Tests
+; Implements NaN triple testing system:
+; - Bit 2: 1 if value is NaN
+; - Bit 1: 1 if quiet NaN
+; - Bit 0: 1 if signaling NaN
+;
+; Triple values:
+; 0b000 (0): Not a NaN
+; 0b101 (5): Signaling NaN
+; 0b110 (6): Quiet NaN
+;
+; ASSUMPTION: All input pointers (edx) are valid and non-null
+
+; Macro: CHECK_NAN_TRIPLE_32
+; Checks 32-bit float NaN classification and returns triple in EAX
+; Input: 32-bit float value in xmm0
+; Output: NaN triple in EAX (bits 2:0)
+%macro CHECK_NAN_TRIPLE_32 0
+  push ecx
+  push esi
+  push edx
+  
+  xor eax, eax
+  ucomiss xmm0, xmm0
+  setp al
+  mov ecx, eax
+  shl ecx, 2
+  
+  ; Extract and check quiet bit (bit 22)
+  movd edx, xmm0
+  and edx, 0x00400000
+  mov esi, edx
+  shr esi, 22
+  and esi, eax
+  and esi, 1
+  shl esi, 1
+  add ecx, esi
+  
+  ; Check for signaling NaN (NaN but not quiet)
+  test edx, edx
+  sete dl
+  and dl, al
+  movzx eax, dl
+  or eax, ecx
+  
+  pop edx
+  pop esi
+  pop ecx
+%endmacro
+
+; Macro: CHECK_NAN_TRIPLE_64
+; Checks 64-bit double NaN classification and returns triple in EAX
+; Input: 64-bit double value should be pre-stored at [edx] by caller
+; Output: NaN triple in EAX (bits 2:0)
+%macro CHECK_NAN_TRIPLE_64 0
+  push ebx
+  push esi
+  sub esp, 12
+  
+  ; Load 64-bit double and use SSE for NaN comparison
+  movsd xmm0, qword [edx]
+  xor eax, eax
+  ucomisd xmm0, xmm0
+  setp al
+  
+  mov ecx, eax
+  movsd qword [esp], xmm0
+  mov edx, 524288
+  and edx, [esp + 4]
+  shl ecx, 2
+  
+  ; Extract quiet bit (bit 51)
+  mov ebx, edx
+  shr ebx, 19
+  and bl, al
+  movzx esi, bl
+  lea ecx, [ecx + 2*esi]
+  
+  ; Check for signaling NaN (NaN but not quiet)
+  test edx, edx
+  sete dl
+  and dl, al
+  movzx eax, dl
+  or eax, ecx
+  
+  add esp, 12
+  pop esi
+  pop ebx
+%endmacro
+
+; Macro: CHECK_NAN_TRIPLE_80
+; Checks 80-bit extended precision NaN classification and returns triple in EAX
+; Input: 80-bit extended precision value in memory at [eax] (10 bytes)
+; Output: NaN triple in EAX (bits 2:0)
+%macro CHECK_NAN_TRIPLE_80 0
+  push ebx
+  push esi
+  sub esp, 20
+  
+  ; Load the 80-bit value and store copy for bit manipulation
+  fld tword [eax]
+  fld st0
+  fstp tword [esp]
+  
+  ; Get bits 63:32 from stored significand
+  mov ecx, [esp + 4]
+  xor eax, eax
+  fucomip st0
+  setp al
+  mov edx, eax
+  shl edx, 2
+  
+  ; Extract quiet bit (bit 30 in high dword)
+  mov ebx, ecx
+  shr ebx, 30
+  and bl, al
+  movzx esi, bl
+  lea edx, [edx + 2*esi]
+  
+  ; Check for signaling NaN using bt instruction
+  bt ecx, 30
+  setae cl
+  and cl, al
+  movzx eax, cl
+  or eax, edx
+  
+  add esp, 20
+  pop esi
+  pop ebx
+%endmacro

unittests/32Bit_ASM/VEX/vgather_qpd_128bit_1xdisp_overflow.asm

@@ -0,0 +1,69 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "XMM1":  ["0", "0", "0", "0"],
+    "XMM7": ["0x2db6b85f6135a9d7", "0xe273f1177f80d2ec", "0", "0"]
+  },
+  "Mode": "32BIT",
+  "Env": { "FEX_HOSTFEATURES" : "enableavx" },
+  "HostFeatures": ["AVX"]
+}
+%endif
+
+; 128-bit
+; 1x displacement
+; 64-bit indexes
+
+; Calculate an address that heavily overflows
+lea eax, [rel .data_mid]
+
+mov ebx, -1
+sub ebx, eax
+mov [rel .index_overflow + 0 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 8
+mov [rel .index_overflow + 1 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 16
+mov [rel .index_overflow + 2 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+add bx, 16
+mov [rel .index_overflow + 3 * 8], ebx
+
+; Calculate new base which offsets from the overflow
+lea eax, [rel .data_mid]
+shl eax, 1
+
+vmovapd ymm7, [rel .data]
+
+vmovaps xmm0, [rel .index_overflow]
+vmovaps xmm1, [rel .mask_1111]
+vgatherqpd xmm7, [xmm0 * 1 + eax + 1], xmm1
+
+hlt
+
+align 32
+
+.mask_1111:
+dq 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000
+
+; Indexing is a signed 64-bit integer.
+.index_overflow:
+dq 0, 0, 0, 0
+
+; Random data, 512-byte per line
+.data:
+db 0x1d, 0x30, 0x09, 0x22, 0x56, 0xa2, 0xcd, 0xf1, 0xb7, 0x62, 0x91, 0x40, 0x67, 0x07, 0x35, 0x0f, 0x59, 0x33, 0x63, 0x52, 0x26, 0xd2, 0x2f, 0x00, 0x8a, 0x36, 0xb7, 0xf7, 0xaf, 0x4f, 0xa1, 0xc0, 0xe3, 0xbe, 0x99, 0x67, 0x5f, 0x80, 0x60, 0xcb, 0x43, 0xfa, 0x5b, 0x86, 0xb1, 0x11, 0xbc, 0xb3, 0x7b, 0x43, 0x5b, 0x45, 0x9e, 0x33, 0x89, 0xb5, 0x1b, 0xb9, 0x33, 0x4f, 0xdb, 0x5d, 0x93, 0xd6, 0x4f, 0xbc, 0x37, 0xde, 0xeb, 0xdb, 0x43, 0x2b, 0x05, 0x60, 0xb8, 0x98, 0x5c, 0xa3, 0xe3, 0x1b, 0x33, 0x03, 0x29, 0x4b, 0x12, 0x4c, 0x1e, 0xe6, 0x5e, 0x0e, 0x6c, 0xa1, 0xb9, 0x36, 0xfa, 0x6c, 0x7f, 0xc6, 0xa8, 0x38, 0x73, 0x2a, 0x0a, 0x25, 0x69, 0xa5, 0x97, 0x3f, 0x24, 0x00, 0x30, 0x4d, 0x27, 0xb3, 0x94, 0x48, 0xef, 0x47, 0x98, 0x71, 0x0d, 0x56, 0x76, 0xec, 0x41, 0x12, 0x9b, 0x7b, 0x9c, 0xf5, 0x85, 0x07, 0x2d, 0x6b, 0xc6, 0xc1, 0x2e, 0x72, 0x22, 0x5a, 0x43, 0xff, 0x1e, 0xec, 0x67, 0x2b, 0x31, 0x96, 0x14, 0x2c, 0xb1, 0x5f, 0x5d, 0x0c, 0xc9, 0xad, 0x15, 0x5f, 0xab, 0x66, 0x14, 0x1c, 0x72, 0xfa, 0x23, 0xef, 0x9f, 0x77, 0xf6, 0x50, 0xb0, 0x70, 0xb8, 0x3c, 0x85, 0x9e, 0x90, 0x69, 0x17, 0x25, 0xae, 0x6e, 0xe2, 0x16, 0x7d, 0x42, 0x38, 0xdf, 0x74, 0x72, 0x7b, 0x97, 0xa9, 0x9e, 0x40, 0x24, 0x85, 0xdc, 0x64, 0xfa, 0xb1, 0x8b, 0x95, 0xe6, 0xe4, 0x13, 0x72, 0xf1, 0x52, 0x2f, 0xa0, 0xd6, 0x52, 0xc0, 0x11, 0xa7, 0xfe, 0xd5, 0x3b, 0x56, 0xca, 0xbc, 0x01, 0xce, 0x3d, 0xd2, 0x30, 0x97, 0x1d, 0xdc, 0xeb, 0x9d, 0xa9, 0x3e, 0x09, 0xef, 0xee, 0x7f, 0x09, 0x7b, 0x82, 0x43, 0x15, 0x2e, 0xa4, 0x2e, 0x97, 0x21, 0x92, 0x7e, 0x69, 0x21, 0x25, 0xda, 0x46, 0x7c, 0x0c, 0xcd, 0x1d, 0xde, 0x42, 0x11, 0xa2, 0xef, 0xa2, 0xc8, 0x32, 0x9a, 0x82, 0xcf, 0x72, 0x7e, 0x22, 0xa6, 0x11, 0xfa, 0xec, 0x0b, 0x77, 0x99, 0x38, 0x03, 0xf6, 0x80, 0xba, 0xea, 0x75, 0x19, 0xb0, 0x48, 0x02, 0xb2, 0x6b, 0xc0, 0x8c, 0xfb, 0xfe, 0xaf, 0x94, 0x4f, 0x6f, 0xb4, 0xcb, 0x1c, 0x27, 0xf0, 0x41, 0xb6, 0x46, 0x41, 0x68, 0x3d, 0x05, 0x79, 0x6b, 0xcd, 0xb7, 0x20, 0xdc, 0x40, 0x81, 0x58, 0xcb, 0x33, 0xa3, 0xf3, 0x34, 0xdc, 0x63, 0x2d, 0xa5, 0xb5, 0xa1, 0xd1, 0xfd, 0x49, 0x5b, 0x46, 0x94, 0x01, 0xa8, 0xf2, 0xd8, 0x93, 0x2c, 0xbb, 0x57, 0xfe, 0x7c, 0x77, 0x3b, 0x19, 0x6f, 0x3c, 0xaa, 0x23, 0x5b, 0xc0, 0xe7, 0x00, 0x41, 0x97, 0x91, 0xe8, 0x00, 0x12, 0xdf, 0xf6, 0x5c, 0x2e, 0xc6, 0x8e, 0xc6, 0x77, 0x59, 0x78, 0x9b, 0xef, 0x63, 0xb0, 0xd7, 0xbb, 0xc4, 0x0b, 0x60, 0x65, 0x3f, 0xfe, 0xbf, 0x04, 0x3e, 0xae, 0xc2, 0xa5, 0x90, 0xe1, 0x2a, 0x56, 0x3f, 0x4c, 0x3f, 0x7a, 0x7d, 0xda, 0x81, 0x50, 0xea, 0x4c, 0xfe, 0xc3, 0xf8, 0x5c, 0x2b, 0x67, 0xb3, 0x9f, 0x8b, 0x95, 0xda, 0x6f, 0x5d, 0xdd, 0x82, 0x7f, 0x52, 0xa2, 0xcc, 0x57, 0xec, 0xc4, 0x14, 0xd2, 0x4f, 0x1b, 0xcb, 0xea, 0xaf, 0x0e, 0x0f, 0x53, 0xaa, 0x56, 0x63, 0xea, 0x36, 0xa6, 0x89, 0x1a, 0x66, 0xc0, 0x4e, 0xf4, 0x1e, 0x02, 0x43, 0xde, 0xde, 0xc8, 0x9e, 0x88, 0x6e, 0x32, 0xd4, 0xcb, 0x47, 0x24, 0x7c, 0x28, 0x38, 0xd4, 0x95, 0xb6, 0xa3, 0x91, 0x69, 0xc7, 0x8d, 0xfd, 0x15, 0xf5, 0xbf, 0xb1, 0x98, 0x8c, 0x57, 0x51, 0xbf, 0x83, 0x6a, 0x35, 0x10, 0x03, 0x50, 0xe5, 0xf7, 0xfa, 0xf8, 0xa5, 0xb0, 0xdb, 0xfb, 0x42, 0x93, 0xbb, 0x17, 0xf7, 0x36, 0xbe, 0x26, 0x66, 0x61, 0xe2
+
+db 0xaf, 0xe0, 0x22, 0x25, 0x2a, 0xae, 0x78, 0x67, 0x8f, 0x7e, 0x9e, 0x59, 0xd7, 0xa3, 0x71, 0xcc, 0x43, 0x85, 0x09, 0xf9, 0x18, 0x52, 0x7b, 0x01, 0x73, 0xcb, 0x31, 0x18, 0x66, 0x79, 0x67, 0x10, 0x67, 0xd8, 0xdf, 0x43, 0xaf, 0x2d, 0x9a, 0x09, 0x9c, 0xd1, 0x37, 0x7e, 0xf5, 0x1c, 0x3c, 0x4f, 0x15, 0xe1, 0x6f, 0xfd, 0x13, 0x3d, 0x53, 0x81, 0xa9, 0x93, 0x5f, 0x92, 0x41, 0x48, 0xec, 0x87, 0x87, 0x1d, 0x0b, 0xaa, 0xaa, 0xd3, 0xc2, 0x98, 0x20, 0xce, 0x28, 0xaf, 0x9d, 0x84, 0x69, 0x4a, 0xfd, 0xc0, 0x9c, 0x2e, 0x50, 0x20, 0xb2, 0x00, 0xc1, 0x81, 0x2a, 0x32, 0x8e, 0x95, 0x20, 0xa7, 0xca, 0x39, 0x28, 0x12, 0x23, 0x0e, 0x43, 0xd3, 0x82, 0x76, 0x73, 0x3c, 0xbf, 0xa9, 0x98, 0xf6, 0x39, 0x6d, 0xd9, 0x15, 0x33, 0x1e, 0x07, 0x7c, 0x08, 0x12, 0x23, 0xbd, 0xd3, 0x34, 0x2d, 0x9a, 0x23, 0x21, 0x46, 0xf3, 0x9a, 0x04, 0x25, 0x62, 0xeb, 0x7e, 0x9a, 0xaa, 0xb6, 0x26, 0xaa, 0x85, 0x01, 0x3a, 0xd8, 0xfc, 0x57, 0x98, 0xb9, 0xe4, 0xc4, 0xe9, 0x11, 0x3e, 0x22, 0x95, 0x3b, 0x41, 0x2b, 0x02, 0x04, 0x6c, 0x75, 0xa5, 0xf2, 0xaa, 0x09, 0x9e, 0x6f, 0xab, 0x1d, 0x2a, 0x5c, 0xde, 0x21, 0xb1, 0x96, 0x2d, 0x86, 0x3f, 0xd0, 0x07, 0x18, 0x1f, 0x87, 0xc2, 0x8f, 0xdf, 0x6a, 0x57, 0x6d, 0x3f, 0x80, 0xc5, 0x08, 0x19, 0xa5, 0x09, 0x65, 0x3d, 0xdc, 0x9e, 0x80, 0x3c, 0x2a, 0x0e, 0x7a, 0x40, 0x04, 0x0b, 0xcc, 0x61, 0xdb, 0x73, 0xfc, 0xa5, 0x0a, 0x42, 0x18, 0xc1, 0xd5, 0xbd, 0x18, 0x78, 0xa1, 0xe4, 0xde, 0x44, 0xec, 0x79, 0xb0, 0x27, 0xaa, 0x45, 0x21, 0x57, 0x19, 0x75, 0x09, 0x5c, 0x58, 0xd5, 0xb9, 0x6f, 0x3b, 0x48, 0x59, 0x41, 0x3e, 0xfd, 0x17, 0x43, 0x27, 0xc3, 0x8d, 0x76, 0x8e, 0x38, 0x47, 0xe7, 0xd2, 0xea, 0x54, 0x73, 0x8a, 0x65, 0x4c, 0x49, 0x91, 0xaf, 0x29, 0x65, 0x0d, 0x81, 0xa4, 0x77, 0xd7, 0x32, 0xd0, 0x69, 0xd9, 0x6b, 0xa3, 0x9b, 0x24, 0xd6, 0x0a, 0xd2, 0x77, 0x38, 0x59, 0x0b, 0xc8, 0x5c, 0xc7, 0x0b, 0x1d, 0xd1, 0xfa, 0xa7, 0x45, 0x3c, 0xeb, 0x5c, 0x8e, 0x25, 0x35, 0x81, 0x6d, 0x6d, 0xfe, 0xb4, 0x63, 0x89, 0xe4, 0xf0, 0xa8, 0xda, 0xb7, 0xd4, 0xff, 0x5d, 0x28, 0x97, 0x11, 0xf9, 0x8d, 0xab, 0x29, 0xd5, 0xd3, 0x1c, 0x70, 0x20, 0x4c, 0x41, 0x16, 0x42, 0xfd, 0xfc, 0x62, 0x82, 0x40, 0x59, 0x34, 0x28, 0xd0, 0xd5, 0xfc, 0xac, 0x97, 0xb8, 0x82, 0x0e, 0x4b, 0xae, 0x51, 0x28, 0x1a, 0xf1, 0x87, 0xd3, 0x20, 0xa3, 0xe7, 0x74, 0x69, 0x3c, 0x54, 0x8d, 0xc5, 0x56, 0x1d, 0xcd, 0x75, 0xae, 0x88, 0x17, 0x30, 0xdf, 0x46, 0x4a, 0xbc, 0x64, 0xff, 0xa2, 0xc1, 0x62, 0xbd, 0x88, 0x7b, 0x3e, 0xa1, 0x0c, 0xa9, 0x13, 0x0e, 0xc1, 0xb4, 0x24, 0xe6, 0x96, 0x1b, 0x9c, 0x9b, 0xac, 0x44, 0x33, 0x5b, 0xda, 0xd5, 0x88, 0x4d, 0xfe, 0x81, 0x09, 0x07, 0x17, 0xcf, 0x14, 0x05, 0xaf, 0xf8, 0x72, 0x14, 0x49, 0x5f, 0x06, 0x62, 0xab, 0xe0, 0x42, 0x70, 0x12, 0x59, 0x41, 0x0f, 0x18, 0x83, 0x68, 0x6d, 0xc6, 0x3c, 0xea, 0xe0, 0x6d, 0xd4, 0xae, 0xa6, 0xf1, 0x63, 0x21, 0x7f, 0xb5, 0x9d, 0x22, 0xf4, 0xd2, 0x49, 0x49, 0xed, 0x07, 0xb1, 0x11, 0xf9, 0x2e, 0x74, 0xbe, 0x35, 0x47, 0xdc, 0xef, 0x85, 0x0b, 0x4d, 0x46, 0xe6, 0x1f, 0x60, 0x6a, 0xa1, 0x8a, 0x4d, 0x46, 0x87, 0x30, 0x8e, 0x9a, 0xba, 0x97, 0x3e, 0x15, 0xb7, 0x33, 0x76, 0x81, 0x69, 0xdb, 0x82, 0x5e, 0xe6, 0x7b, 0xec, 0xd2, 0x80, 0x7f, 0x17, 0xf1, 0x73, 0xe2
+
+.data_mid:
+db 0xd7, 0xa9, 0x35, 0x61, 0x5f, 0xb8, 0xb6, 0x2d, 0x29, 0x34, 0x63, 0xbf, 0xe2, 0x1c, 0x34, 0xe9, 0xf5, 0xff, 0x34, 0x8a, 0x2f, 0xea, 0xd4, 0x3f, 0x3b, 0xfe, 0x6e, 0xdf, 0xa6, 0xd8, 0xc6, 0xb8, 0xc5, 0xff, 0x12, 0x97, 0x53, 0xda, 0x86, 0xa8, 0x0a, 0x12, 0x4e, 0x5d, 0x96, 0x65, 0x51, 0x22, 0xe2, 0x9d, 0x08, 0x71, 0x84, 0x19, 0x8b, 0xbf, 0x29, 0xd3, 0x3f, 0xab, 0xde, 0xe4, 0x27, 0x8b, 0x99, 0xcc, 0xb1, 0x7c, 0xa5, 0x71, 0x91, 0x9a, 0x0b, 0xad, 0x75, 0x86, 0xe3, 0x9c, 0x4e, 0x0c, 0x01, 0xb3, 0x12, 0x33, 0x90, 0x81, 0x7c, 0x71, 0x2c, 0x70, 0x61, 0xd5, 0x39, 0x0c, 0x45, 0xfc, 0x27, 0xaf, 0xbb, 0xd9, 0x26, 0x1b, 0x33, 0xb4, 0x0d, 0xf8, 0xd6, 0x2d, 0x09, 0xc7, 0x8c, 0xbf, 0x48, 0x53, 0x14, 0x94, 0x76, 0x25, 0xc7, 0x0c, 0x69, 0x49, 0x82, 0xb4, 0x2f, 0x48, 0x38, 0x44, 0x9d, 0x90, 0x6d, 0x66, 0x35, 0xe9, 0x3e, 0x2f, 0x2a, 0xb7, 0xe1, 0xb1, 0x2b, 0x99, 0x08, 0x6f, 0x5c, 0x6c, 0xdf, 0xdb, 0x10, 0xe2, 0xaa, 0x86, 0xe7, 0xf8, 0x9e, 0x62, 0xde, 0xa5, 0x81, 0x6b, 0x20, 0x47, 0xa9, 0x06, 0x49, 0xc0, 0x78, 0x8c, 0x70, 0x93, 0x7e, 0xda, 0xda, 0x5e, 0x3b, 0x23, 0xf9, 0xcc, 0x87, 0xdf, 0x48, 0x4f, 0xd6, 0x77, 0xce, 0x45, 0xe1, 0xdc, 0x0c, 0x7a, 0x0c, 0x50, 0x15, 0x63, 0x8c, 0x48, 0xd3, 0x8e, 0xfa, 0xcc, 0xac, 0x1a, 0x83, 0xde, 0xb1, 0x87, 0x2a, 0x58, 0x5c, 0xa5, 0x20, 0x3d, 0xaa, 0x1e, 0x5d, 0x71, 0xa6, 0x57, 0x75, 0x82, 0xb7, 0x33, 0x9e, 0x6b, 0xf3, 0x35, 0x02, 0x98, 0x03, 0xe1, 0x3b, 0xd2, 0x9f, 0x7a, 0x06, 0x85, 0xef, 0x7d, 0xd9, 0xf2, 0x0c, 0x9e, 0xce, 0xb9, 0xce, 0x13, 0x4a, 0x9e, 0x8a, 0x29, 0xe6, 0xe5, 0xe4, 0x39, 0xba, 0xfd, 0xa3, 0x33, 0xa8, 0x13, 0x9e, 0xa5, 0x11, 0x37, 0x69, 0xbc, 0xda, 0x11, 0x49, 0x2d, 0x4a, 0xef, 0x20, 0x8b, 0x7a, 0xb8, 0x9c, 0xc3, 0xaf, 0x26, 0x71, 0xd9, 0xa2, 0xf6, 0x0f, 0x85, 0x87, 0xa8, 0x6c, 0xf9, 0x99, 0xa2, 0xb2, 0x36, 0x2d, 0x78, 0x10, 0xe4, 0x33, 0x8d, 0xa4, 0x63, 0xea, 0x02, 0xb9, 0xac, 0x2f, 0x90, 0x39, 0x2d, 0x0e, 0x2e, 0xf5, 0x08, 0xa5, 0x5c, 0x8e, 0x71, 0x30, 0x0d, 0x1b, 0x84, 0x7a, 0xd7, 0xd4, 0xab, 0x81, 0x82, 0x18, 0x37, 0xf3, 0x28, 0x6f, 0x4e, 0x28, 0x71, 0xda, 0xc9, 0x99, 0x46, 0x14, 0x46, 0x77, 0x01, 0x16, 0x21, 0xae, 0x83, 0x93, 0x86, 0x7f, 0x5a, 0xee, 0xd5, 0xdf, 0x48, 0x5b, 0x15, 0xc8, 0x09, 0x30, 0x8f, 0x01, 0xcc, 0x95, 0x30, 0xd9, 0xf7, 0x72, 0x97, 0xfd, 0x9d, 0xec, 0x9f, 0xbf, 0x5c, 0xbf, 0x4f, 0xca, 0x33, 0xb4, 0xd2, 0xa2, 0xb9, 0x08, 0x9c, 0x40, 0x25, 0x3f, 0x86, 0xdc, 0x83, 0x70, 0x2f, 0xfb, 0x2a, 0xf8, 0x61, 0x1f, 0xa1, 0x1f, 0x36, 0x04, 0xe2, 0xef, 0x1c, 0xa4, 0xcd, 0x3c, 0x7f, 0xc5, 0x73, 0x9c, 0x2e, 0xeb, 0x03, 0x79, 0xd1, 0x02, 0xfc, 0x6f, 0xbd, 0x5a, 0x95, 0xb2, 0xf6, 0x25, 0x96, 0xe6, 0x80, 0x0a, 0xc5, 0xc7, 0xca, 0x8d, 0x31, 0xae, 0xf0, 0x49, 0xcf, 0x43, 0x06, 0x27, 0x7f, 0x25, 0xc7, 0x4c, 0xb7, 0xfc, 0x73, 0xd3, 0x04, 0xd3, 0xb9, 0x9f, 0x74, 0xed, 0x9e, 0x3c, 0xf0, 0xcf, 0x26, 0x2b, 0xd9, 0xcb, 0x78, 0x2a, 0xef, 0x72, 0xf7, 0xb6, 0x78, 0x30, 0x2d, 0x8c, 0x83, 0x73, 0x66, 0x74, 0x3d, 0x66, 0x0a, 0x74, 0x5a, 0x3f, 0x9f, 0x6e, 0x56, 0x68, 0x01, 0xc2, 0xca, 0x2b, 0xa1, 0x25, 0x36, 0x9c, 0x3b, 0xa4, 0x5e, 0x44, 0xf1, 0x18, 0x1d, 0xb6, 0x1a, 0x3a, 0xee, 0x8d, 0x67, 0x34, 0x9c
+
+db 0xdd, 0x48, 0x14, 0xc2, 0x5f, 0xd8, 0xe5, 0x71, 0x22, 0xbf, 0xbc, 0x84, 0xda, 0xc1, 0xb1, 0x22, 0x55, 0xa4, 0x63, 0x41, 0x77, 0xac, 0x40, 0x2d, 0x44, 0x73, 0x8c, 0x14, 0xba, 0x5e, 0x63, 0x68, 0x65, 0x61, 0x6d, 0xec, 0xe2, 0x6d, 0x37, 0x22, 0x04, 0xeb, 0xc7, 0xd4, 0xc9, 0x62, 0x56, 0x13, 0x96, 0x29, 0x03, 0xf4, 0x55, 0xe2, 0x58, 0x7d, 0xda, 0x52, 0x2e, 0x94, 0x07, 0xe6, 0xef, 0xc0, 0xee, 0x9e, 0x0b, 0xf7, 0xcd, 0x13, 0x8b, 0x7d, 0xea, 0xdc, 0xf8, 0xf1, 0xcb, 0xad, 0x49, 0x97, 0xc9, 0x98, 0x0b, 0xcf, 0x84, 0x8e, 0x8e, 0xbb, 0x06, 0x2e, 0x54, 0xf5, 0xa7, 0xbd, 0x70, 0x7e, 0x38, 0x69, 0x8d, 0xb0, 0x01, 0x7b, 0x41, 0x80, 0x09, 0x44, 0xfd, 0x7e, 0x21, 0xb4, 0xbe, 0x6b, 0x4a, 0xb7, 0xca, 0x2d, 0x19, 0xfe, 0x6d, 0xd6, 0x11, 0x29, 0xbb, 0xb2, 0x16, 0xf1, 0xe7, 0x92, 0x71, 0xda, 0x7e, 0x68, 0x3a, 0xe0, 0xea, 0x89, 0x8d, 0xe0, 0x44, 0x48, 0x25, 0x92, 0x37, 0x54, 0x26, 0xf2, 0xab, 0xb3, 0x3b, 0xdb, 0xbb, 0x2b, 0x5c, 0xf5, 0xbc, 0xc7, 0x97, 0xdb, 0xc7, 0x49, 0x25, 0x7c, 0xc2, 0x80, 0x02, 0x69, 0xd4, 0xda, 0xda, 0xe1, 0x04, 0xf3, 0x19, 0xb8, 0xc9, 0xb2, 0xfb, 0x1e, 0x47, 0xa9, 0x0c, 0xa3, 0x48, 0xce, 0xc2, 0x9e, 0x3b, 0x28, 0x23, 0x5a, 0x20, 0x44, 0x77, 0x40, 0xe2, 0xd7, 0x20, 0xd5, 0x71, 0x6f, 0xd4, 0x3c, 0x68, 0x38, 0x9b, 0x89, 0x2e, 0x2d, 0xa8, 0x1f, 0x99, 0xb5, 0x8a, 0x66, 0x07, 0x59, 0x75, 0x9e, 0xf8, 0xd9, 0xbe, 0x85, 0x6a, 0x20, 0x92, 0x9d, 0xd2, 0x5e, 0x45, 0xc0, 0x60, 0xbe, 0x85, 0x0b, 0x84, 0x47, 0xf5, 0xa8, 0x43, 0x87, 0xf1, 0x21, 0x21, 0xb0, 0x3b, 0x04, 0x13, 0x16, 0x3e, 0xdf, 0xc3, 0xc6, 0x04, 0x73, 0xcd, 0x92, 0x76, 0xfb, 0xe7, 0x9c, 0xd3, 0x46, 0x11, 0x78, 0xca, 0x12, 0xd9, 0x4a, 0x35, 0xf1, 0x6e, 0x89, 0x8b, 0xe9, 0x7a, 0x04, 0xba, 0x18, 0x25, 0x7c, 0x9e, 0xe6, 0x4f, 0xc2, 0x56, 0x05, 0x72, 0xc3, 0x76, 0xee, 0x7d, 0x77, 0x19, 0x7a, 0x73, 0x2c, 0x81, 0xb8, 0xc7, 0xd9, 0x7f, 0x17, 0x5d, 0x30, 0xda, 0x77, 0x3c, 0x14, 0x88, 0xe8, 0xe4, 0xbf, 0xee, 0x21, 0x1c, 0x29, 0x4e, 0x58, 0xa8, 0x8a, 0x5c, 0xae, 0xa2, 0x1c, 0x7c, 0x25, 0x7c, 0x1c, 0x39, 0xa4, 0x28, 0x4b, 0x78, 0x52, 0xae, 0x2c, 0xbb, 0x5f, 0xbf, 0x51, 0x09, 0x20, 0x76, 0xb2, 0x7d, 0xb1, 0x63, 0x84, 0xc5, 0x49, 0x8a, 0x73, 0xdb, 0x76, 0x1d, 0x25, 0x31, 0xf2, 0x1e, 0x19, 0x38, 0xc8, 0x3b, 0x51, 0x3c, 0x13, 0x52, 0x84, 0xae, 0xc2, 0xe4, 0x8a, 0x57, 0x0d, 0xde, 0x8d, 0x18, 0x48, 0x9a, 0xbd, 0xbf, 0xf3, 0xea, 0x79, 0x17, 0x06, 0x96, 0x72, 0x08, 0x60, 0x95, 0xf9, 0x6f, 0x25, 0x0c, 0xb7, 0x9d, 0x98, 0x23, 0x01, 0xc8, 0x7a, 0xdb, 0x75, 0x63, 0x64, 0x14, 0x5e, 0x10, 0xf5, 0x16, 0x48, 0xbc, 0xc6, 0x7e, 0x24, 0xf3, 0xad, 0x57, 0x3f, 0x7d, 0x6c, 0xab, 0x18, 0x8c, 0x12, 0xc5, 0x0c, 0xd8, 0xb5, 0x1e, 0x43, 0x7c, 0x23, 0x17, 0x48, 0xba, 0x76, 0x3b, 0xd9, 0x2b, 0xae, 0x1b, 0xef, 0x58, 0xfa, 0x87, 0xad, 0x9b, 0x6d, 0xf9, 0xab, 0xa8, 0x3c, 0xfc, 0x59, 0x67, 0xa6, 0x2c, 0xc7, 0x75, 0xa4, 0x97, 0xca, 0x18, 0x18, 0x04, 0x2c, 0xb3, 0x0e, 0xa9, 0x69, 0x33, 0x67, 0xa2, 0xc6, 0xbc, 0x98, 0x48, 0x71, 0x11, 0x05, 0x30, 0xf6, 0xa9, 0x61, 0x40, 0x46, 0xf1, 0x41, 0x37, 0xd0, 0x6b, 0x7c, 0x1f, 0x03, 0x5c, 0xe9, 0xf4, 0x59, 0x1d, 0x35, 0xf0, 0x98, 0x42, 0x4a, 0x92, 0x2a, 0xc3, 0x9a, 0xb8, 0xa5

unittests/32Bit_ASM/VEX/vgather_qpd_128bit_2xdisp_overflow.asm

@@ -0,0 +1,73 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "XMM1":  ["0", "0", "0", "0"],
+    "XMM7": ["0x2db6b85f6135a9d7", "0xe273f1177f80d2ec", "0", "0"]
+  },
+  "Mode": "32BIT",
+  "Env": { "FEX_HOSTFEATURES" : "enableavx" },
+  "HostFeatures": ["AVX"]
+}
+%endif
+
+; 128-bit
+; 2x displacement
+; 64-bit indexes
+
+; Calculate an address that heavily overflows
+lea eax, [rel .data_mid]
+
+mov ebx, -1
+sub ebx, eax
+sar ebx, 1
+mov [rel .index_overflow + 0 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 8
+sar ebx, 1
+mov [rel .index_overflow + 1 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 16
+sar ebx, 1
+mov [rel .index_overflow + 2 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+add bx, 16
+sar ebx, 1
+mov [rel .index_overflow + 3 * 8], ebx
+
+; Calculate new base which offsets from the overflow
+lea eax, [rel .data_mid]
+shl eax, 1
+
+vmovapd ymm7, [rel .data]
+
+vmovaps xmm0, [rel .index_overflow]
+vmovaps xmm1, [rel .mask_1111]
+vgatherqpd xmm7, [xmm0 * 2 + eax + 2], xmm1
+
+hlt
+
+align 32
+
+.mask_1111:
+dq 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000
+
+; Indexing is a signed 64-bit integer.
+.index_overflow:
+dq 0, 0, 0, 0
+
+; Random data, 512-byte per line
+.data:
+db 0x1d, 0x30, 0x09, 0x22, 0x56, 0xa2, 0xcd, 0xf1, 0xb7, 0x62, 0x91, 0x40, 0x67, 0x07, 0x35, 0x0f, 0x59, 0x33, 0x63, 0x52, 0x26, 0xd2, 0x2f, 0x00, 0x8a, 0x36, 0xb7, 0xf7, 0xaf, 0x4f, 0xa1, 0xc0, 0xe3, 0xbe, 0x99, 0x67, 0x5f, 0x80, 0x60, 0xcb, 0x43, 0xfa, 0x5b, 0x86, 0xb1, 0x11, 0xbc, 0xb3, 0x7b, 0x43, 0x5b, 0x45, 0x9e, 0x33, 0x89, 0xb5, 0x1b, 0xb9, 0x33, 0x4f, 0xdb, 0x5d, 0x93, 0xd6, 0x4f, 0xbc, 0x37, 0xde, 0xeb, 0xdb, 0x43, 0x2b, 0x05, 0x60, 0xb8, 0x98, 0x5c, 0xa3, 0xe3, 0x1b, 0x33, 0x03, 0x29, 0x4b, 0x12, 0x4c, 0x1e, 0xe6, 0x5e, 0x0e, 0x6c, 0xa1, 0xb9, 0x36, 0xfa, 0x6c, 0x7f, 0xc6, 0xa8, 0x38, 0x73, 0x2a, 0x0a, 0x25, 0x69, 0xa5, 0x97, 0x3f, 0x24, 0x00, 0x30, 0x4d, 0x27, 0xb3, 0x94, 0x48, 0xef, 0x47, 0x98, 0x71, 0x0d, 0x56, 0x76, 0xec, 0x41, 0x12, 0x9b, 0x7b, 0x9c, 0xf5, 0x85, 0x07, 0x2d, 0x6b, 0xc6, 0xc1, 0x2e, 0x72, 0x22, 0x5a, 0x43, 0xff, 0x1e, 0xec, 0x67, 0x2b, 0x31, 0x96, 0x14, 0x2c, 0xb1, 0x5f, 0x5d, 0x0c, 0xc9, 0xad, 0x15, 0x5f, 0xab, 0x66, 0x14, 0x1c, 0x72, 0xfa, 0x23, 0xef, 0x9f, 0x77, 0xf6, 0x50, 0xb0, 0x70, 0xb8, 0x3c, 0x85, 0x9e, 0x90, 0x69, 0x17, 0x25, 0xae, 0x6e, 0xe2, 0x16, 0x7d, 0x42, 0x38, 0xdf, 0x74, 0x72, 0x7b, 0x97, 0xa9, 0x9e, 0x40, 0x24, 0x85, 0xdc, 0x64, 0xfa, 0xb1, 0x8b, 0x95, 0xe6, 0xe4, 0x13, 0x72, 0xf1, 0x52, 0x2f, 0xa0, 0xd6, 0x52, 0xc0, 0x11, 0xa7, 0xfe, 0xd5, 0x3b, 0x56, 0xca, 0xbc, 0x01, 0xce, 0x3d, 0xd2, 0x30, 0x97, 0x1d, 0xdc, 0xeb, 0x9d, 0xa9, 0x3e, 0x09, 0xef, 0xee, 0x7f, 0x09, 0x7b, 0x82, 0x43, 0x15, 0x2e, 0xa4, 0x2e, 0x97, 0x21, 0x92, 0x7e, 0x69, 0x21, 0x25, 0xda, 0x46, 0x7c, 0x0c, 0xcd, 0x1d, 0xde, 0x42, 0x11, 0xa2, 0xef, 0xa2, 0xc8, 0x32, 0x9a, 0x82, 0xcf, 0x72, 0x7e, 0x22, 0xa6, 0x11, 0xfa, 0xec, 0x0b, 0x77, 0x99, 0x38, 0x03, 0xf6, 0x80, 0xba, 0xea, 0x75, 0x19, 0xb0, 0x48, 0x02, 0xb2, 0x6b, 0xc0, 0x8c, 0xfb, 0xfe, 0xaf, 0x94, 0x4f, 0x6f, 0xb4, 0xcb, 0x1c, 0x27, 0xf0, 0x41, 0xb6, 0x46, 0x41, 0x68, 0x3d, 0x05, 0x79, 0x6b, 0xcd, 0xb7, 0x20, 0xdc, 0x40, 0x81, 0x58, 0xcb, 0x33, 0xa3, 0xf3, 0x34, 0xdc, 0x63, 0x2d, 0xa5, 0xb5, 0xa1, 0xd1, 0xfd, 0x49, 0x5b, 0x46, 0x94, 0x01, 0xa8, 0xf2, 0xd8, 0x93, 0x2c, 0xbb, 0x57, 0xfe, 0x7c, 0x77, 0x3b, 0x19, 0x6f, 0x3c, 0xaa, 0x23, 0x5b, 0xc0, 0xe7, 0x00, 0x41, 0x97, 0x91, 0xe8, 0x00, 0x12, 0xdf, 0xf6, 0x5c, 0x2e, 0xc6, 0x8e, 0xc6, 0x77, 0x59, 0x78, 0x9b, 0xef, 0x63, 0xb0, 0xd7, 0xbb, 0xc4, 0x0b, 0x60, 0x65, 0x3f, 0xfe, 0xbf, 0x04, 0x3e, 0xae, 0xc2, 0xa5, 0x90, 0xe1, 0x2a, 0x56, 0x3f, 0x4c, 0x3f, 0x7a, 0x7d, 0xda, 0x81, 0x50, 0xea, 0x4c, 0xfe, 0xc3, 0xf8, 0x5c, 0x2b, 0x67, 0xb3, 0x9f, 0x8b, 0x95, 0xda, 0x6f, 0x5d, 0xdd, 0x82, 0x7f, 0x52, 0xa2, 0xcc, 0x57, 0xec, 0xc4, 0x14, 0xd2, 0x4f, 0x1b, 0xcb, 0xea, 0xaf, 0x0e, 0x0f, 0x53, 0xaa, 0x56, 0x63, 0xea, 0x36, 0xa6, 0x89, 0x1a, 0x66, 0xc0, 0x4e, 0xf4, 0x1e, 0x02, 0x43, 0xde, 0xde, 0xc8, 0x9e, 0x88, 0x6e, 0x32, 0xd4, 0xcb, 0x47, 0x24, 0x7c, 0x28, 0x38, 0xd4, 0x95, 0xb6, 0xa3, 0x91, 0x69, 0xc7, 0x8d, 0xfd, 0x15, 0xf5, 0xbf, 0xb1, 0x98, 0x8c, 0x57, 0x51, 0xbf, 0x83, 0x6a, 0x35, 0x10, 0x03, 0x50, 0xe5, 0xf7, 0xfa, 0xf8, 0xa5, 0xb0, 0xdb, 0xfb, 0x42, 0x93, 0xbb, 0x17, 0xf7, 0x36, 0xbe, 0x26, 0x66, 0x61, 0xe2
+
+db 0xaf, 0xe0, 0x22, 0x25, 0x2a, 0xae, 0x78, 0x67, 0x8f, 0x7e, 0x9e, 0x59, 0xd7, 0xa3, 0x71, 0xcc, 0x43, 0x85, 0x09, 0xf9, 0x18, 0x52, 0x7b, 0x01, 0x73, 0xcb, 0x31, 0x18, 0x66, 0x79, 0x67, 0x10, 0x67, 0xd8, 0xdf, 0x43, 0xaf, 0x2d, 0x9a, 0x09, 0x9c, 0xd1, 0x37, 0x7e, 0xf5, 0x1c, 0x3c, 0x4f, 0x15, 0xe1, 0x6f, 0xfd, 0x13, 0x3d, 0x53, 0x81, 0xa9, 0x93, 0x5f, 0x92, 0x41, 0x48, 0xec, 0x87, 0x87, 0x1d, 0x0b, 0xaa, 0xaa, 0xd3, 0xc2, 0x98, 0x20, 0xce, 0x28, 0xaf, 0x9d, 0x84, 0x69, 0x4a, 0xfd, 0xc0, 0x9c, 0x2e, 0x50, 0x20, 0xb2, 0x00, 0xc1, 0x81, 0x2a, 0x32, 0x8e, 0x95, 0x20, 0xa7, 0xca, 0x39, 0x28, 0x12, 0x23, 0x0e, 0x43, 0xd3, 0x82, 0x76, 0x73, 0x3c, 0xbf, 0xa9, 0x98, 0xf6, 0x39, 0x6d, 0xd9, 0x15, 0x33, 0x1e, 0x07, 0x7c, 0x08, 0x12, 0x23, 0xbd, 0xd3, 0x34, 0x2d, 0x9a, 0x23, 0x21, 0x46, 0xf3, 0x9a, 0x04, 0x25, 0x62, 0xeb, 0x7e, 0x9a, 0xaa, 0xb6, 0x26, 0xaa, 0x85, 0x01, 0x3a, 0xd8, 0xfc, 0x57, 0x98, 0xb9, 0xe4, 0xc4, 0xe9, 0x11, 0x3e, 0x22, 0x95, 0x3b, 0x41, 0x2b, 0x02, 0x04, 0x6c, 0x75, 0xa5, 0xf2, 0xaa, 0x09, 0x9e, 0x6f, 0xab, 0x1d, 0x2a, 0x5c, 0xde, 0x21, 0xb1, 0x96, 0x2d, 0x86, 0x3f, 0xd0, 0x07, 0x18, 0x1f, 0x87, 0xc2, 0x8f, 0xdf, 0x6a, 0x57, 0x6d, 0x3f, 0x80, 0xc5, 0x08, 0x19, 0xa5, 0x09, 0x65, 0x3d, 0xdc, 0x9e, 0x80, 0x3c, 0x2a, 0x0e, 0x7a, 0x40, 0x04, 0x0b, 0xcc, 0x61, 0xdb, 0x73, 0xfc, 0xa5, 0x0a, 0x42, 0x18, 0xc1, 0xd5, 0xbd, 0x18, 0x78, 0xa1, 0xe4, 0xde, 0x44, 0xec, 0x79, 0xb0, 0x27, 0xaa, 0x45, 0x21, 0x57, 0x19, 0x75, 0x09, 0x5c, 0x58, 0xd5, 0xb9, 0x6f, 0x3b, 0x48, 0x59, 0x41, 0x3e, 0xfd, 0x17, 0x43, 0x27, 0xc3, 0x8d, 0x76, 0x8e, 0x38, 0x47, 0xe7, 0xd2, 0xea, 0x54, 0x73, 0x8a, 0x65, 0x4c, 0x49, 0x91, 0xaf, 0x29, 0x65, 0x0d, 0x81, 0xa4, 0x77, 0xd7, 0x32, 0xd0, 0x69, 0xd9, 0x6b, 0xa3, 0x9b, 0x24, 0xd6, 0x0a, 0xd2, 0x77, 0x38, 0x59, 0x0b, 0xc8, 0x5c, 0xc7, 0x0b, 0x1d, 0xd1, 0xfa, 0xa7, 0x45, 0x3c, 0xeb, 0x5c, 0x8e, 0x25, 0x35, 0x81, 0x6d, 0x6d, 0xfe, 0xb4, 0x63, 0x89, 0xe4, 0xf0, 0xa8, 0xda, 0xb7, 0xd4, 0xff, 0x5d, 0x28, 0x97, 0x11, 0xf9, 0x8d, 0xab, 0x29, 0xd5, 0xd3, 0x1c, 0x70, 0x20, 0x4c, 0x41, 0x16, 0x42, 0xfd, 0xfc, 0x62, 0x82, 0x40, 0x59, 0x34, 0x28, 0xd0, 0xd5, 0xfc, 0xac, 0x97, 0xb8, 0x82, 0x0e, 0x4b, 0xae, 0x51, 0x28, 0x1a, 0xf1, 0x87, 0xd3, 0x20, 0xa3, 0xe7, 0x74, 0x69, 0x3c, 0x54, 0x8d, 0xc5, 0x56, 0x1d, 0xcd, 0x75, 0xae, 0x88, 0x17, 0x30, 0xdf, 0x46, 0x4a, 0xbc, 0x64, 0xff, 0xa2, 0xc1, 0x62, 0xbd, 0x88, 0x7b, 0x3e, 0xa1, 0x0c, 0xa9, 0x13, 0x0e, 0xc1, 0xb4, 0x24, 0xe6, 0x96, 0x1b, 0x9c, 0x9b, 0xac, 0x44, 0x33, 0x5b, 0xda, 0xd5, 0x88, 0x4d, 0xfe, 0x81, 0x09, 0x07, 0x17, 0xcf, 0x14, 0x05, 0xaf, 0xf8, 0x72, 0x14, 0x49, 0x5f, 0x06, 0x62, 0xab, 0xe0, 0x42, 0x70, 0x12, 0x59, 0x41, 0x0f, 0x18, 0x83, 0x68, 0x6d, 0xc6, 0x3c, 0xea, 0xe0, 0x6d, 0xd4, 0xae, 0xa6, 0xf1, 0x63, 0x21, 0x7f, 0xb5, 0x9d, 0x22, 0xf4, 0xd2, 0x49, 0x49, 0xed, 0x07, 0xb1, 0x11, 0xf9, 0x2e, 0x74, 0xbe, 0x35, 0x47, 0xdc, 0xef, 0x85, 0x0b, 0x4d, 0x46, 0xe6, 0x1f, 0x60, 0x6a, 0xa1, 0x8a, 0x4d, 0x46, 0x87, 0x30, 0x8e, 0x9a, 0xba, 0x97, 0x3e, 0x15, 0xb7, 0x33, 0x76, 0x81, 0x69, 0xdb, 0x82, 0x5e, 0xe6, 0x7b, 0xec, 0xd2, 0x80, 0x7f, 0x17, 0xf1, 0x73, 0xe2
+
+.data_mid:
+db 0xd7, 0xa9, 0x35, 0x61, 0x5f, 0xb8, 0xb6, 0x2d, 0x29, 0x34, 0x63, 0xbf, 0xe2, 0x1c, 0x34, 0xe9, 0xf5, 0xff, 0x34, 0x8a, 0x2f, 0xea, 0xd4, 0x3f, 0x3b, 0xfe, 0x6e, 0xdf, 0xa6, 0xd8, 0xc6, 0xb8, 0xc5, 0xff, 0x12, 0x97, 0x53, 0xda, 0x86, 0xa8, 0x0a, 0x12, 0x4e, 0x5d, 0x96, 0x65, 0x51, 0x22, 0xe2, 0x9d, 0x08, 0x71, 0x84, 0x19, 0x8b, 0xbf, 0x29, 0xd3, 0x3f, 0xab, 0xde, 0xe4, 0x27, 0x8b, 0x99, 0xcc, 0xb1, 0x7c, 0xa5, 0x71, 0x91, 0x9a, 0x0b, 0xad, 0x75, 0x86, 0xe3, 0x9c, 0x4e, 0x0c, 0x01, 0xb3, 0x12, 0x33, 0x90, 0x81, 0x7c, 0x71, 0x2c, 0x70, 0x61, 0xd5, 0x39, 0x0c, 0x45, 0xfc, 0x27, 0xaf, 0xbb, 0xd9, 0x26, 0x1b, 0x33, 0xb4, 0x0d, 0xf8, 0xd6, 0x2d, 0x09, 0xc7, 0x8c, 0xbf, 0x48, 0x53, 0x14, 0x94, 0x76, 0x25, 0xc7, 0x0c, 0x69, 0x49, 0x82, 0xb4, 0x2f, 0x48, 0x38, 0x44, 0x9d, 0x90, 0x6d, 0x66, 0x35, 0xe9, 0x3e, 0x2f, 0x2a, 0xb7, 0xe1, 0xb1, 0x2b, 0x99, 0x08, 0x6f, 0x5c, 0x6c, 0xdf, 0xdb, 0x10, 0xe2, 0xaa, 0x86, 0xe7, 0xf8, 0x9e, 0x62, 0xde, 0xa5, 0x81, 0x6b, 0x20, 0x47, 0xa9, 0x06, 0x49, 0xc0, 0x78, 0x8c, 0x70, 0x93, 0x7e, 0xda, 0xda, 0x5e, 0x3b, 0x23, 0xf9, 0xcc, 0x87, 0xdf, 0x48, 0x4f, 0xd6, 0x77, 0xce, 0x45, 0xe1, 0xdc, 0x0c, 0x7a, 0x0c, 0x50, 0x15, 0x63, 0x8c, 0x48, 0xd3, 0x8e, 0xfa, 0xcc, 0xac, 0x1a, 0x83, 0xde, 0xb1, 0x87, 0x2a, 0x58, 0x5c, 0xa5, 0x20, 0x3d, 0xaa, 0x1e, 0x5d, 0x71, 0xa6, 0x57, 0x75, 0x82, 0xb7, 0x33, 0x9e, 0x6b, 0xf3, 0x35, 0x02, 0x98, 0x03, 0xe1, 0x3b, 0xd2, 0x9f, 0x7a, 0x06, 0x85, 0xef, 0x7d, 0xd9, 0xf2, 0x0c, 0x9e, 0xce, 0xb9, 0xce, 0x13, 0x4a, 0x9e, 0x8a, 0x29, 0xe6, 0xe5, 0xe4, 0x39, 0xba, 0xfd, 0xa3, 0x33, 0xa8, 0x13, 0x9e, 0xa5, 0x11, 0x37, 0x69, 0xbc, 0xda, 0x11, 0x49, 0x2d, 0x4a, 0xef, 0x20, 0x8b, 0x7a, 0xb8, 0x9c, 0xc3, 0xaf, 0x26, 0x71, 0xd9, 0xa2, 0xf6, 0x0f, 0x85, 0x87, 0xa8, 0x6c, 0xf9, 0x99, 0xa2, 0xb2, 0x36, 0x2d, 0x78, 0x10, 0xe4, 0x33, 0x8d, 0xa4, 0x63, 0xea, 0x02, 0xb9, 0xac, 0x2f, 0x90, 0x39, 0x2d, 0x0e, 0x2e, 0xf5, 0x08, 0xa5, 0x5c, 0x8e, 0x71, 0x30, 0x0d, 0x1b, 0x84, 0x7a, 0xd7, 0xd4, 0xab, 0x81, 0x82, 0x18, 0x37, 0xf3, 0x28, 0x6f, 0x4e, 0x28, 0x71, 0xda, 0xc9, 0x99, 0x46, 0x14, 0x46, 0x77, 0x01, 0x16, 0x21, 0xae, 0x83, 0x93, 0x86, 0x7f, 0x5a, 0xee, 0xd5, 0xdf, 0x48, 0x5b, 0x15, 0xc8, 0x09, 0x30, 0x8f, 0x01, 0xcc, 0x95, 0x30, 0xd9, 0xf7, 0x72, 0x97, 0xfd, 0x9d, 0xec, 0x9f, 0xbf, 0x5c, 0xbf, 0x4f, 0xca, 0x33, 0xb4, 0xd2, 0xa2, 0xb9, 0x08, 0x9c, 0x40, 0x25, 0x3f, 0x86, 0xdc, 0x83, 0x70, 0x2f, 0xfb, 0x2a, 0xf8, 0x61, 0x1f, 0xa1, 0x1f, 0x36, 0x04, 0xe2, 0xef, 0x1c, 0xa4, 0xcd, 0x3c, 0x7f, 0xc5, 0x73, 0x9c, 0x2e, 0xeb, 0x03, 0x79, 0xd1, 0x02, 0xfc, 0x6f, 0xbd, 0x5a, 0x95, 0xb2, 0xf6, 0x25, 0x96, 0xe6, 0x80, 0x0a, 0xc5, 0xc7, 0xca, 0x8d, 0x31, 0xae, 0xf0, 0x49, 0xcf, 0x43, 0x06, 0x27, 0x7f, 0x25, 0xc7, 0x4c, 0xb7, 0xfc, 0x73, 0xd3, 0x04, 0xd3, 0xb9, 0x9f, 0x74, 0xed, 0x9e, 0x3c, 0xf0, 0xcf, 0x26, 0x2b, 0xd9, 0xcb, 0x78, 0x2a, 0xef, 0x72, 0xf7, 0xb6, 0x78, 0x30, 0x2d, 0x8c, 0x83, 0x73, 0x66, 0x74, 0x3d, 0x66, 0x0a, 0x74, 0x5a, 0x3f, 0x9f, 0x6e, 0x56, 0x68, 0x01, 0xc2, 0xca, 0x2b, 0xa1, 0x25, 0x36, 0x9c, 0x3b, 0xa4, 0x5e, 0x44, 0xf1, 0x18, 0x1d, 0xb6, 0x1a, 0x3a, 0xee, 0x8d, 0x67, 0x34, 0x9c
+
+db 0xdd, 0x48, 0x14, 0xc2, 0x5f, 0xd8, 0xe5, 0x71, 0x22, 0xbf, 0xbc, 0x84, 0xda, 0xc1, 0xb1, 0x22, 0x55, 0xa4, 0x63, 0x41, 0x77, 0xac, 0x40, 0x2d, 0x44, 0x73, 0x8c, 0x14, 0xba, 0x5e, 0x63, 0x68, 0x65, 0x61, 0x6d, 0xec, 0xe2, 0x6d, 0x37, 0x22, 0x04, 0xeb, 0xc7, 0xd4, 0xc9, 0x62, 0x56, 0x13, 0x96, 0x29, 0x03, 0xf4, 0x55, 0xe2, 0x58, 0x7d, 0xda, 0x52, 0x2e, 0x94, 0x07, 0xe6, 0xef, 0xc0, 0xee, 0x9e, 0x0b, 0xf7, 0xcd, 0x13, 0x8b, 0x7d, 0xea, 0xdc, 0xf8, 0xf1, 0xcb, 0xad, 0x49, 0x97, 0xc9, 0x98, 0x0b, 0xcf, 0x84, 0x8e, 0x8e, 0xbb, 0x06, 0x2e, 0x54, 0xf5, 0xa7, 0xbd, 0x70, 0x7e, 0x38, 0x69, 0x8d, 0xb0, 0x01, 0x7b, 0x41, 0x80, 0x09, 0x44, 0xfd, 0x7e, 0x21, 0xb4, 0xbe, 0x6b, 0x4a, 0xb7, 0xca, 0x2d, 0x19, 0xfe, 0x6d, 0xd6, 0x11, 0x29, 0xbb, 0xb2, 0x16, 0xf1, 0xe7, 0x92, 0x71, 0xda, 0x7e, 0x68, 0x3a, 0xe0, 0xea, 0x89, 0x8d, 0xe0, 0x44, 0x48, 0x25, 0x92, 0x37, 0x54, 0x26, 0xf2, 0xab, 0xb3, 0x3b, 0xdb, 0xbb, 0x2b, 0x5c, 0xf5, 0xbc, 0xc7, 0x97, 0xdb, 0xc7, 0x49, 0x25, 0x7c, 0xc2, 0x80, 0x02, 0x69, 0xd4, 0xda, 0xda, 0xe1, 0x04, 0xf3, 0x19, 0xb8, 0xc9, 0xb2, 0xfb, 0x1e, 0x47, 0xa9, 0x0c, 0xa3, 0x48, 0xce, 0xc2, 0x9e, 0x3b, 0x28, 0x23, 0x5a, 0x20, 0x44, 0x77, 0x40, 0xe2, 0xd7, 0x20, 0xd5, 0x71, 0x6f, 0xd4, 0x3c, 0x68, 0x38, 0x9b, 0x89, 0x2e, 0x2d, 0xa8, 0x1f, 0x99, 0xb5, 0x8a, 0x66, 0x07, 0x59, 0x75, 0x9e, 0xf8, 0xd9, 0xbe, 0x85, 0x6a, 0x20, 0x92, 0x9d, 0xd2, 0x5e, 0x45, 0xc0, 0x60, 0xbe, 0x85, 0x0b, 0x84, 0x47, 0xf5, 0xa8, 0x43, 0x87, 0xf1, 0x21, 0x21, 0xb0, 0x3b, 0x04, 0x13, 0x16, 0x3e, 0xdf, 0xc3, 0xc6, 0x04, 0x73, 0xcd, 0x92, 0x76, 0xfb, 0xe7, 0x9c, 0xd3, 0x46, 0x11, 0x78, 0xca, 0x12, 0xd9, 0x4a, 0x35, 0xf1, 0x6e, 0x89, 0x8b, 0xe9, 0x7a, 0x04, 0xba, 0x18, 0x25, 0x7c, 0x9e, 0xe6, 0x4f, 0xc2, 0x56, 0x05, 0x72, 0xc3, 0x76, 0xee, 0x7d, 0x77, 0x19, 0x7a, 0x73, 0x2c, 0x81, 0xb8, 0xc7, 0xd9, 0x7f, 0x17, 0x5d, 0x30, 0xda, 0x77, 0x3c, 0x14, 0x88, 0xe8, 0xe4, 0xbf, 0xee, 0x21, 0x1c, 0x29, 0x4e, 0x58, 0xa8, 0x8a, 0x5c, 0xae, 0xa2, 0x1c, 0x7c, 0x25, 0x7c, 0x1c, 0x39, 0xa4, 0x28, 0x4b, 0x78, 0x52, 0xae, 0x2c, 0xbb, 0x5f, 0xbf, 0x51, 0x09, 0x20, 0x76, 0xb2, 0x7d, 0xb1, 0x63, 0x84, 0xc5, 0x49, 0x8a, 0x73, 0xdb, 0x76, 0x1d, 0x25, 0x31, 0xf2, 0x1e, 0x19, 0x38, 0xc8, 0x3b, 0x51, 0x3c, 0x13, 0x52, 0x84, 0xae, 0xc2, 0xe4, 0x8a, 0x57, 0x0d, 0xde, 0x8d, 0x18, 0x48, 0x9a, 0xbd, 0xbf, 0xf3, 0xea, 0x79, 0x17, 0x06, 0x96, 0x72, 0x08, 0x60, 0x95, 0xf9, 0x6f, 0x25, 0x0c, 0xb7, 0x9d, 0x98, 0x23, 0x01, 0xc8, 0x7a, 0xdb, 0x75, 0x63, 0x64, 0x14, 0x5e, 0x10, 0xf5, 0x16, 0x48, 0xbc, 0xc6, 0x7e, 0x24, 0xf3, 0xad, 0x57, 0x3f, 0x7d, 0x6c, 0xab, 0x18, 0x8c, 0x12, 0xc5, 0x0c, 0xd8, 0xb5, 0x1e, 0x43, 0x7c, 0x23, 0x17, 0x48, 0xba, 0x76, 0x3b, 0xd9, 0x2b, 0xae, 0x1b, 0xef, 0x58, 0xfa, 0x87, 0xad, 0x9b, 0x6d, 0xf9, 0xab, 0xa8, 0x3c, 0xfc, 0x59, 0x67, 0xa6, 0x2c, 0xc7, 0x75, 0xa4, 0x97, 0xca, 0x18, 0x18, 0x04, 0x2c, 0xb3, 0x0e, 0xa9, 0x69, 0x33, 0x67, 0xa2, 0xc6, 0xbc, 0x98, 0x48, 0x71, 0x11, 0x05, 0x30, 0xf6, 0xa9, 0x61, 0x40, 0x46, 0xf1, 0x41, 0x37, 0xd0, 0x6b, 0x7c, 0x1f, 0x03, 0x5c, 0xe9, 0xf4, 0x59, 0x1d, 0x35, 0xf0, 0x98, 0x42, 0x4a, 0x92, 0x2a, 0xc3, 0x9a, 0xb8, 0xa5

unittests/32Bit_ASM/VEX/vgather_qpd_128bit_4xdisp_overflow.asm

@@ -0,0 +1,73 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "XMM1":  ["0", "0", "0", "0"],
+    "XMM7": ["0x2db6b85f6135a9d7", "0xe273f1177f80d2ec", "0", "0"]
+  },
+  "Mode": "32BIT",
+  "Env": { "FEX_HOSTFEATURES" : "enableavx" },
+  "HostFeatures": ["AVX"]
+}
+%endif
+
+; 128-bit
+; 4x displacement
+; 64-bit indexes
+
+; Calculate an address that heavily overflows
+lea eax, [rel .data_mid]
+
+mov ebx, -1
+sub ebx, eax
+sar ebx, 2
+mov [rel .index_overflow + 0 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 8
+sar ebx, 2
+mov [rel .index_overflow + 1 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 16
+sar ebx, 2
+mov [rel .index_overflow + 2 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+add bx, 16
+sar ebx, 2
+mov [rel .index_overflow + 3 * 8], ebx
+
+; Calculate new base which offsets from the overflow
+lea eax, [rel .data_mid]
+shl eax, 1
+
+vmovapd ymm7, [rel .data]
+
+vmovaps xmm0, [rel .index_overflow]
+vmovaps xmm1, [rel .mask_1111]
+vgatherqpd xmm7, [xmm0 * 4 + eax + 4], xmm1
+
+hlt
+
+align 32
+
+.mask_1111:
+dq 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000
+
+; Indexing is a signed 64-bit integer.
+.index_overflow:
+dq 0, 0, 0, 0
+
+; Random data, 512-byte per line
+.data:
+db 0x1d, 0x30, 0x09, 0x22, 0x56, 0xa2, 0xcd, 0xf1, 0xb7, 0x62, 0x91, 0x40, 0x67, 0x07, 0x35, 0x0f, 0x59, 0x33, 0x63, 0x52, 0x26, 0xd2, 0x2f, 0x00, 0x8a, 0x36, 0xb7, 0xf7, 0xaf, 0x4f, 0xa1, 0xc0, 0xe3, 0xbe, 0x99, 0x67, 0x5f, 0x80, 0x60, 0xcb, 0x43, 0xfa, 0x5b, 0x86, 0xb1, 0x11, 0xbc, 0xb3, 0x7b, 0x43, 0x5b, 0x45, 0x9e, 0x33, 0x89, 0xb5, 0x1b, 0xb9, 0x33, 0x4f, 0xdb, 0x5d, 0x93, 0xd6, 0x4f, 0xbc, 0x37, 0xde, 0xeb, 0xdb, 0x43, 0x2b, 0x05, 0x60, 0xb8, 0x98, 0x5c, 0xa3, 0xe3, 0x1b, 0x33, 0x03, 0x29, 0x4b, 0x12, 0x4c, 0x1e, 0xe6, 0x5e, 0x0e, 0x6c, 0xa1, 0xb9, 0x36, 0xfa, 0x6c, 0x7f, 0xc6, 0xa8, 0x38, 0x73, 0x2a, 0x0a, 0x25, 0x69, 0xa5, 0x97, 0x3f, 0x24, 0x00, 0x30, 0x4d, 0x27, 0xb3, 0x94, 0x48, 0xef, 0x47, 0x98, 0x71, 0x0d, 0x56, 0x76, 0xec, 0x41, 0x12, 0x9b, 0x7b, 0x9c, 0xf5, 0x85, 0x07, 0x2d, 0x6b, 0xc6, 0xc1, 0x2e, 0x72, 0x22, 0x5a, 0x43, 0xff, 0x1e, 0xec, 0x67, 0x2b, 0x31, 0x96, 0x14, 0x2c, 0xb1, 0x5f, 0x5d, 0x0c, 0xc9, 0xad, 0x15, 0x5f, 0xab, 0x66, 0x14, 0x1c, 0x72, 0xfa, 0x23, 0xef, 0x9f, 0x77, 0xf6, 0x50, 0xb0, 0x70, 0xb8, 0x3c, 0x85, 0x9e, 0x90, 0x69, 0x17, 0x25, 0xae, 0x6e, 0xe2, 0x16, 0x7d, 0x42, 0x38, 0xdf, 0x74, 0x72, 0x7b, 0x97, 0xa9, 0x9e, 0x40, 0x24, 0x85, 0xdc, 0x64, 0xfa, 0xb1, 0x8b, 0x95, 0xe6, 0xe4, 0x13, 0x72, 0xf1, 0x52, 0x2f, 0xa0, 0xd6, 0x52, 0xc0, 0x11, 0xa7, 0xfe, 0xd5, 0x3b, 0x56, 0xca, 0xbc, 0x01, 0xce, 0x3d, 0xd2, 0x30, 0x97, 0x1d, 0xdc, 0xeb, 0x9d, 0xa9, 0x3e, 0x09, 0xef, 0xee, 0x7f, 0x09, 0x7b, 0x82, 0x43, 0x15, 0x2e, 0xa4, 0x2e, 0x97, 0x21, 0x92, 0x7e, 0x69, 0x21, 0x25, 0xda, 0x46, 0x7c, 0x0c, 0xcd, 0x1d, 0xde, 0x42, 0x11, 0xa2, 0xef, 0xa2, 0xc8, 0x32, 0x9a, 0x82, 0xcf, 0x72, 0x7e, 0x22, 0xa6, 0x11, 0xfa, 0xec, 0x0b, 0x77, 0x99, 0x38, 0x03, 0xf6, 0x80, 0xba, 0xea, 0x75, 0x19, 0xb0, 0x48, 0x02, 0xb2, 0x6b, 0xc0, 0x8c, 0xfb, 0xfe, 0xaf, 0x94, 0x4f, 0x6f, 0xb4, 0xcb, 0x1c, 0x27, 0xf0, 0x41, 0xb6, 0x46, 0x41, 0x68, 0x3d, 0x05, 0x79, 0x6b, 0xcd, 0xb7, 0x20, 0xdc, 0x40, 0x81, 0x58, 0xcb, 0x33, 0xa3, 0xf3, 0x34, 0xdc, 0x63, 0x2d, 0xa5, 0xb5, 0xa1, 0xd1, 0xfd, 0x49, 0x5b, 0x46, 0x94, 0x01, 0xa8, 0xf2, 0xd8, 0x93, 0x2c, 0xbb, 0x57, 0xfe, 0x7c, 0x77, 0x3b, 0x19, 0x6f, 0x3c, 0xaa, 0x23, 0x5b, 0xc0, 0xe7, 0x00, 0x41, 0x97, 0x91, 0xe8, 0x00, 0x12, 0xdf, 0xf6, 0x5c, 0x2e, 0xc6, 0x8e, 0xc6, 0x77, 0x59, 0x78, 0x9b, 0xef, 0x63, 0xb0, 0xd7, 0xbb, 0xc4, 0x0b, 0x60, 0x65, 0x3f, 0xfe, 0xbf, 0x04, 0x3e, 0xae, 0xc2, 0xa5, 0x90, 0xe1, 0x2a, 0x56, 0x3f, 0x4c, 0x3f, 0x7a, 0x7d, 0xda, 0x81, 0x50, 0xea, 0x4c, 0xfe, 0xc3, 0xf8, 0x5c, 0x2b, 0x67, 0xb3, 0x9f, 0x8b, 0x95, 0xda, 0x6f, 0x5d, 0xdd, 0x82, 0x7f, 0x52, 0xa2, 0xcc, 0x57, 0xec, 0xc4, 0x14, 0xd2, 0x4f, 0x1b, 0xcb, 0xea, 0xaf, 0x0e, 0x0f, 0x53, 0xaa, 0x56, 0x63, 0xea, 0x36, 0xa6, 0x89, 0x1a, 0x66, 0xc0, 0x4e, 0xf4, 0x1e, 0x02, 0x43, 0xde, 0xde, 0xc8, 0x9e, 0x88, 0x6e, 0x32, 0xd4, 0xcb, 0x47, 0x24, 0x7c, 0x28, 0x38, 0xd4, 0x95, 0xb6, 0xa3, 0x91, 0x69, 0xc7, 0x8d, 0xfd, 0x15, 0xf5, 0xbf, 0xb1, 0x98, 0x8c, 0x57, 0x51, 0xbf, 0x83, 0x6a, 0x35, 0x10, 0x03, 0x50, 0xe5, 0xf7, 0xfa, 0xf8, 0xa5, 0xb0, 0xdb, 0xfb, 0x42, 0x93, 0xbb, 0x17, 0xf7, 0x36, 0xbe, 0x26, 0x66, 0x61, 0xe2
+
+db 0xaf, 0xe0, 0x22, 0x25, 0x2a, 0xae, 0x78, 0x67, 0x8f, 0x7e, 0x9e, 0x59, 0xd7, 0xa3, 0x71, 0xcc, 0x43, 0x85, 0x09, 0xf9, 0x18, 0x52, 0x7b, 0x01, 0x73, 0xcb, 0x31, 0x18, 0x66, 0x79, 0x67, 0x10, 0x67, 0xd8, 0xdf, 0x43, 0xaf, 0x2d, 0x9a, 0x09, 0x9c, 0xd1, 0x37, 0x7e, 0xf5, 0x1c, 0x3c, 0x4f, 0x15, 0xe1, 0x6f, 0xfd, 0x13, 0x3d, 0x53, 0x81, 0xa9, 0x93, 0x5f, 0x92, 0x41, 0x48, 0xec, 0x87, 0x87, 0x1d, 0x0b, 0xaa, 0xaa, 0xd3, 0xc2, 0x98, 0x20, 0xce, 0x28, 0xaf, 0x9d, 0x84, 0x69, 0x4a, 0xfd, 0xc0, 0x9c, 0x2e, 0x50, 0x20, 0xb2, 0x00, 0xc1, 0x81, 0x2a, 0x32, 0x8e, 0x95, 0x20, 0xa7, 0xca, 0x39, 0x28, 0x12, 0x23, 0x0e, 0x43, 0xd3, 0x82, 0x76, 0x73, 0x3c, 0xbf, 0xa9, 0x98, 0xf6, 0x39, 0x6d, 0xd9, 0x15, 0x33, 0x1e, 0x07, 0x7c, 0x08, 0x12, 0x23, 0xbd, 0xd3, 0x34, 0x2d, 0x9a, 0x23, 0x21, 0x46, 0xf3, 0x9a, 0x04, 0x25, 0x62, 0xeb, 0x7e, 0x9a, 0xaa, 0xb6, 0x26, 0xaa, 0x85, 0x01, 0x3a, 0xd8, 0xfc, 0x57, 0x98, 0xb9, 0xe4, 0xc4, 0xe9, 0x11, 0x3e, 0x22, 0x95, 0x3b, 0x41, 0x2b, 0x02, 0x04, 0x6c, 0x75, 0xa5, 0xf2, 0xaa, 0x09, 0x9e, 0x6f, 0xab, 0x1d, 0x2a, 0x5c, 0xde, 0x21, 0xb1, 0x96, 0x2d, 0x86, 0x3f, 0xd0, 0x07, 0x18, 0x1f, 0x87, 0xc2, 0x8f, 0xdf, 0x6a, 0x57, 0x6d, 0x3f, 0x80, 0xc5, 0x08, 0x19, 0xa5, 0x09, 0x65, 0x3d, 0xdc, 0x9e, 0x80, 0x3c, 0x2a, 0x0e, 0x7a, 0x40, 0x04, 0x0b, 0xcc, 0x61, 0xdb, 0x73, 0xfc, 0xa5, 0x0a, 0x42, 0x18, 0xc1, 0xd5, 0xbd, 0x18, 0x78, 0xa1, 0xe4, 0xde, 0x44, 0xec, 0x79, 0xb0, 0x27, 0xaa, 0x45, 0x21, 0x57, 0x19, 0x75, 0x09, 0x5c, 0x58, 0xd5, 0xb9, 0x6f, 0x3b, 0x48, 0x59, 0x41, 0x3e, 0xfd, 0x17, 0x43, 0x27, 0xc3, 0x8d, 0x76, 0x8e, 0x38, 0x47, 0xe7, 0xd2, 0xea, 0x54, 0x73, 0x8a, 0x65, 0x4c, 0x49, 0x91, 0xaf, 0x29, 0x65, 0x0d, 0x81, 0xa4, 0x77, 0xd7, 0x32, 0xd0, 0x69, 0xd9, 0x6b, 0xa3, 0x9b, 0x24, 0xd6, 0x0a, 0xd2, 0x77, 0x38, 0x59, 0x0b, 0xc8, 0x5c, 0xc7, 0x0b, 0x1d, 0xd1, 0xfa, 0xa7, 0x45, 0x3c, 0xeb, 0x5c, 0x8e, 0x25, 0x35, 0x81, 0x6d, 0x6d, 0xfe, 0xb4, 0x63, 0x89, 0xe4, 0xf0, 0xa8, 0xda, 0xb7, 0xd4, 0xff, 0x5d, 0x28, 0x97, 0x11, 0xf9, 0x8d, 0xab, 0x29, 0xd5, 0xd3, 0x1c, 0x70, 0x20, 0x4c, 0x41, 0x16, 0x42, 0xfd, 0xfc, 0x62, 0x82, 0x40, 0x59, 0x34, 0x28, 0xd0, 0xd5, 0xfc, 0xac, 0x97, 0xb8, 0x82, 0x0e, 0x4b, 0xae, 0x51, 0x28, 0x1a, 0xf1, 0x87, 0xd3, 0x20, 0xa3, 0xe7, 0x74, 0x69, 0x3c, 0x54, 0x8d, 0xc5, 0x56, 0x1d, 0xcd, 0x75, 0xae, 0x88, 0x17, 0x30, 0xdf, 0x46, 0x4a, 0xbc, 0x64, 0xff, 0xa2, 0xc1, 0x62, 0xbd, 0x88, 0x7b, 0x3e, 0xa1, 0x0c, 0xa9, 0x13, 0x0e, 0xc1, 0xb4, 0x24, 0xe6, 0x96, 0x1b, 0x9c, 0x9b, 0xac, 0x44, 0x33, 0x5b, 0xda, 0xd5, 0x88, 0x4d, 0xfe, 0x81, 0x09, 0x07, 0x17, 0xcf, 0x14, 0x05, 0xaf, 0xf8, 0x72, 0x14, 0x49, 0x5f, 0x06, 0x62, 0xab, 0xe0, 0x42, 0x70, 0x12, 0x59, 0x41, 0x0f, 0x18, 0x83, 0x68, 0x6d, 0xc6, 0x3c, 0xea, 0xe0, 0x6d, 0xd4, 0xae, 0xa6, 0xf1, 0x63, 0x21, 0x7f, 0xb5, 0x9d, 0x22, 0xf4, 0xd2, 0x49, 0x49, 0xed, 0x07, 0xb1, 0x11, 0xf9, 0x2e, 0x74, 0xbe, 0x35, 0x47, 0xdc, 0xef, 0x85, 0x0b, 0x4d, 0x46, 0xe6, 0x1f, 0x60, 0x6a, 0xa1, 0x8a, 0x4d, 0x46, 0x87, 0x30, 0x8e, 0x9a, 0xba, 0x97, 0x3e, 0x15, 0xb7, 0x33, 0x76, 0x81, 0x69, 0xdb, 0x82, 0x5e, 0xe6, 0x7b, 0xec, 0xd2, 0x80, 0x7f, 0x17, 0xf1, 0x73, 0xe2
+
+.data_mid:
+db 0xd7, 0xa9, 0x35, 0x61, 0x5f, 0xb8, 0xb6, 0x2d, 0x29, 0x34, 0x63, 0xbf, 0xe2, 0x1c, 0x34, 0xe9, 0xf5, 0xff, 0x34, 0x8a, 0x2f, 0xea, 0xd4, 0x3f, 0x3b, 0xfe, 0x6e, 0xdf, 0xa6, 0xd8, 0xc6, 0xb8, 0xc5, 0xff, 0x12, 0x97, 0x53, 0xda, 0x86, 0xa8, 0x0a, 0x12, 0x4e, 0x5d, 0x96, 0x65, 0x51, 0x22, 0xe2, 0x9d, 0x08, 0x71, 0x84, 0x19, 0x8b, 0xbf, 0x29, 0xd3, 0x3f, 0xab, 0xde, 0xe4, 0x27, 0x8b, 0x99, 0xcc, 0xb1, 0x7c, 0xa5, 0x71, 0x91, 0x9a, 0x0b, 0xad, 0x75, 0x86, 0xe3, 0x9c, 0x4e, 0x0c, 0x01, 0xb3, 0x12, 0x33, 0x90, 0x81, 0x7c, 0x71, 0x2c, 0x70, 0x61, 0xd5, 0x39, 0x0c, 0x45, 0xfc, 0x27, 0xaf, 0xbb, 0xd9, 0x26, 0x1b, 0x33, 0xb4, 0x0d, 0xf8, 0xd6, 0x2d, 0x09, 0xc7, 0x8c, 0xbf, 0x48, 0x53, 0x14, 0x94, 0x76, 0x25, 0xc7, 0x0c, 0x69, 0x49, 0x82, 0xb4, 0x2f, 0x48, 0x38, 0x44, 0x9d, 0x90, 0x6d, 0x66, 0x35, 0xe9, 0x3e, 0x2f, 0x2a, 0xb7, 0xe1, 0xb1, 0x2b, 0x99, 0x08, 0x6f, 0x5c, 0x6c, 0xdf, 0xdb, 0x10, 0xe2, 0xaa, 0x86, 0xe7, 0xf8, 0x9e, 0x62, 0xde, 0xa5, 0x81, 0x6b, 0x20, 0x47, 0xa9, 0x06, 0x49, 0xc0, 0x78, 0x8c, 0x70, 0x93, 0x7e, 0xda, 0xda, 0x5e, 0x3b, 0x23, 0xf9, 0xcc, 0x87, 0xdf, 0x48, 0x4f, 0xd6, 0x77, 0xce, 0x45, 0xe1, 0xdc, 0x0c, 0x7a, 0x0c, 0x50, 0x15, 0x63, 0x8c, 0x48, 0xd3, 0x8e, 0xfa, 0xcc, 0xac, 0x1a, 0x83, 0xde, 0xb1, 0x87, 0x2a, 0x58, 0x5c, 0xa5, 0x20, 0x3d, 0xaa, 0x1e, 0x5d, 0x71, 0xa6, 0x57, 0x75, 0x82, 0xb7, 0x33, 0x9e, 0x6b, 0xf3, 0x35, 0x02, 0x98, 0x03, 0xe1, 0x3b, 0xd2, 0x9f, 0x7a, 0x06, 0x85, 0xef, 0x7d, 0xd9, 0xf2, 0x0c, 0x9e, 0xce, 0xb9, 0xce, 0x13, 0x4a, 0x9e, 0x8a, 0x29, 0xe6, 0xe5, 0xe4, 0x39, 0xba, 0xfd, 0xa3, 0x33, 0xa8, 0x13, 0x9e, 0xa5, 0x11, 0x37, 0x69, 0xbc, 0xda, 0x11, 0x49, 0x2d, 0x4a, 0xef, 0x20, 0x8b, 0x7a, 0xb8, 0x9c, 0xc3, 0xaf, 0x26, 0x71, 0xd9, 0xa2, 0xf6, 0x0f, 0x85, 0x87, 0xa8, 0x6c, 0xf9, 0x99, 0xa2, 0xb2, 0x36, 0x2d, 0x78, 0x10, 0xe4, 0x33, 0x8d, 0xa4, 0x63, 0xea, 0x02, 0xb9, 0xac, 0x2f, 0x90, 0x39, 0x2d, 0x0e, 0x2e, 0xf5, 0x08, 0xa5, 0x5c, 0x8e, 0x71, 0x30, 0x0d, 0x1b, 0x84, 0x7a, 0xd7, 0xd4, 0xab, 0x81, 0x82, 0x18, 0x37, 0xf3, 0x28, 0x6f, 0x4e, 0x28, 0x71, 0xda, 0xc9, 0x99, 0x46, 0x14, 0x46, 0x77, 0x01, 0x16, 0x21, 0xae, 0x83, 0x93, 0x86, 0x7f, 0x5a, 0xee, 0xd5, 0xdf, 0x48, 0x5b, 0x15, 0xc8, 0x09, 0x30, 0x8f, 0x01, 0xcc, 0x95, 0x30, 0xd9, 0xf7, 0x72, 0x97, 0xfd, 0x9d, 0xec, 0x9f, 0xbf, 0x5c, 0xbf, 0x4f, 0xca, 0x33, 0xb4, 0xd2, 0xa2, 0xb9, 0x08, 0x9c, 0x40, 0x25, 0x3f, 0x86, 0xdc, 0x83, 0x70, 0x2f, 0xfb, 0x2a, 0xf8, 0x61, 0x1f, 0xa1, 0x1f, 0x36, 0x04, 0xe2, 0xef, 0x1c, 0xa4, 0xcd, 0x3c, 0x7f, 0xc5, 0x73, 0x9c, 0x2e, 0xeb, 0x03, 0x79, 0xd1, 0x02, 0xfc, 0x6f, 0xbd, 0x5a, 0x95, 0xb2, 0xf6, 0x25, 0x96, 0xe6, 0x80, 0x0a, 0xc5, 0xc7, 0xca, 0x8d, 0x31, 0xae, 0xf0, 0x49, 0xcf, 0x43, 0x06, 0x27, 0x7f, 0x25, 0xc7, 0x4c, 0xb7, 0xfc, 0x73, 0xd3, 0x04, 0xd3, 0xb9, 0x9f, 0x74, 0xed, 0x9e, 0x3c, 0xf0, 0xcf, 0x26, 0x2b, 0xd9, 0xcb, 0x78, 0x2a, 0xef, 0x72, 0xf7, 0xb6, 0x78, 0x30, 0x2d, 0x8c, 0x83, 0x73, 0x66, 0x74, 0x3d, 0x66, 0x0a, 0x74, 0x5a, 0x3f, 0x9f, 0x6e, 0x56, 0x68, 0x01, 0xc2, 0xca, 0x2b, 0xa1, 0x25, 0x36, 0x9c, 0x3b, 0xa4, 0x5e, 0x44, 0xf1, 0x18, 0x1d, 0xb6, 0x1a, 0x3a, 0xee, 0x8d, 0x67, 0x34, 0x9c
+
+db 0xdd, 0x48, 0x14, 0xc2, 0x5f, 0xd8, 0xe5, 0x71, 0x22, 0xbf, 0xbc, 0x84, 0xda, 0xc1, 0xb1, 0x22, 0x55, 0xa4, 0x63, 0x41, 0x77, 0xac, 0x40, 0x2d, 0x44, 0x73, 0x8c, 0x14, 0xba, 0x5e, 0x63, 0x68, 0x65, 0x61, 0x6d, 0xec, 0xe2, 0x6d, 0x37, 0x22, 0x04, 0xeb, 0xc7, 0xd4, 0xc9, 0x62, 0x56, 0x13, 0x96, 0x29, 0x03, 0xf4, 0x55, 0xe2, 0x58, 0x7d, 0xda, 0x52, 0x2e, 0x94, 0x07, 0xe6, 0xef, 0xc0, 0xee, 0x9e, 0x0b, 0xf7, 0xcd, 0x13, 0x8b, 0x7d, 0xea, 0xdc, 0xf8, 0xf1, 0xcb, 0xad, 0x49, 0x97, 0xc9, 0x98, 0x0b, 0xcf, 0x84, 0x8e, 0x8e, 0xbb, 0x06, 0x2e, 0x54, 0xf5, 0xa7, 0xbd, 0x70, 0x7e, 0x38, 0x69, 0x8d, 0xb0, 0x01, 0x7b, 0x41, 0x80, 0x09, 0x44, 0xfd, 0x7e, 0x21, 0xb4, 0xbe, 0x6b, 0x4a, 0xb7, 0xca, 0x2d, 0x19, 0xfe, 0x6d, 0xd6, 0x11, 0x29, 0xbb, 0xb2, 0x16, 0xf1, 0xe7, 0x92, 0x71, 0xda, 0x7e, 0x68, 0x3a, 0xe0, 0xea, 0x89, 0x8d, 0xe0, 0x44, 0x48, 0x25, 0x92, 0x37, 0x54, 0x26, 0xf2, 0xab, 0xb3, 0x3b, 0xdb, 0xbb, 0x2b, 0x5c, 0xf5, 0xbc, 0xc7, 0x97, 0xdb, 0xc7, 0x49, 0x25, 0x7c, 0xc2, 0x80, 0x02, 0x69, 0xd4, 0xda, 0xda, 0xe1, 0x04, 0xf3, 0x19, 0xb8, 0xc9, 0xb2, 0xfb, 0x1e, 0x47, 0xa9, 0x0c, 0xa3, 0x48, 0xce, 0xc2, 0x9e, 0x3b, 0x28, 0x23, 0x5a, 0x20, 0x44, 0x77, 0x40, 0xe2, 0xd7, 0x20, 0xd5, 0x71, 0x6f, 0xd4, 0x3c, 0x68, 0x38, 0x9b, 0x89, 0x2e, 0x2d, 0xa8, 0x1f, 0x99, 0xb5, 0x8a, 0x66, 0x07, 0x59, 0x75, 0x9e, 0xf8, 0xd9, 0xbe, 0x85, 0x6a, 0x20, 0x92, 0x9d, 0xd2, 0x5e, 0x45, 0xc0, 0x60, 0xbe, 0x85, 0x0b, 0x84, 0x47, 0xf5, 0xa8, 0x43, 0x87, 0xf1, 0x21, 0x21, 0xb0, 0x3b, 0x04, 0x13, 0x16, 0x3e, 0xdf, 0xc3, 0xc6, 0x04, 0x73, 0xcd, 0x92, 0x76, 0xfb, 0xe7, 0x9c, 0xd3, 0x46, 0x11, 0x78, 0xca, 0x12, 0xd9, 0x4a, 0x35, 0xf1, 0x6e, 0x89, 0x8b, 0xe9, 0x7a, 0x04, 0xba, 0x18, 0x25, 0x7c, 0x9e, 0xe6, 0x4f, 0xc2, 0x56, 0x05, 0x72, 0xc3, 0x76, 0xee, 0x7d, 0x77, 0x19, 0x7a, 0x73, 0x2c, 0x81, 0xb8, 0xc7, 0xd9, 0x7f, 0x17, 0x5d, 0x30, 0xda, 0x77, 0x3c, 0x14, 0x88, 0xe8, 0xe4, 0xbf, 0xee, 0x21, 0x1c, 0x29, 0x4e, 0x58, 0xa8, 0x8a, 0x5c, 0xae, 0xa2, 0x1c, 0x7c, 0x25, 0x7c, 0x1c, 0x39, 0xa4, 0x28, 0x4b, 0x78, 0x52, 0xae, 0x2c, 0xbb, 0x5f, 0xbf, 0x51, 0x09, 0x20, 0x76, 0xb2, 0x7d, 0xb1, 0x63, 0x84, 0xc5, 0x49, 0x8a, 0x73, 0xdb, 0x76, 0x1d, 0x25, 0x31, 0xf2, 0x1e, 0x19, 0x38, 0xc8, 0x3b, 0x51, 0x3c, 0x13, 0x52, 0x84, 0xae, 0xc2, 0xe4, 0x8a, 0x57, 0x0d, 0xde, 0x8d, 0x18, 0x48, 0x9a, 0xbd, 0xbf, 0xf3, 0xea, 0x79, 0x17, 0x06, 0x96, 0x72, 0x08, 0x60, 0x95, 0xf9, 0x6f, 0x25, 0x0c, 0xb7, 0x9d, 0x98, 0x23, 0x01, 0xc8, 0x7a, 0xdb, 0x75, 0x63, 0x64, 0x14, 0x5e, 0x10, 0xf5, 0x16, 0x48, 0xbc, 0xc6, 0x7e, 0x24, 0xf3, 0xad, 0x57, 0x3f, 0x7d, 0x6c, 0xab, 0x18, 0x8c, 0x12, 0xc5, 0x0c, 0xd8, 0xb5, 0x1e, 0x43, 0x7c, 0x23, 0x17, 0x48, 0xba, 0x76, 0x3b, 0xd9, 0x2b, 0xae, 0x1b, 0xef, 0x58, 0xfa, 0x87, 0xad, 0x9b, 0x6d, 0xf9, 0xab, 0xa8, 0x3c, 0xfc, 0x59, 0x67, 0xa6, 0x2c, 0xc7, 0x75, 0xa4, 0x97, 0xca, 0x18, 0x18, 0x04, 0x2c, 0xb3, 0x0e, 0xa9, 0x69, 0x33, 0x67, 0xa2, 0xc6, 0xbc, 0x98, 0x48, 0x71, 0x11, 0x05, 0x30, 0xf6, 0xa9, 0x61, 0x40, 0x46, 0xf1, 0x41, 0x37, 0xd0, 0x6b, 0x7c, 0x1f, 0x03, 0x5c, 0xe9, 0xf4, 0x59, 0x1d, 0x35, 0xf0, 0x98, 0x42, 0x4a, 0x92, 0x2a, 0xc3, 0x9a, 0xb8, 0xa5

unittests/32Bit_ASM/VEX/vgather_qpd_128bit_8xdisp_overflow.asm

@@ -0,0 +1,73 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "XMM1":  ["0", "0", "0", "0"],
+    "XMM7": ["0x2db6b85f6135a9d7", "0xe273f1177f80d2ec", "0", "0"]
+  },
+  "Mode": "32BIT",
+  "Env": { "FEX_HOSTFEATURES" : "enableavx" },
+  "HostFeatures": ["AVX"]
+}
+%endif
+
+; 128-bit
+; 8x displacement
+; 64-bit indexes
+
+; Calculate an address that heavily overflows
+lea eax, [rel .data_mid]
+
+mov ebx, -1
+sub ebx, eax
+sar ebx, 3
+mov [rel .index_overflow + 0 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 8
+sar ebx, 3
+mov [rel .index_overflow + 1 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 16
+sar ebx, 3
+mov [rel .index_overflow + 2 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+add bx, 16
+sar ebx, 3
+mov [rel .index_overflow + 3 * 8], ebx
+
+; Calculate new base which offsets from the overflow
+lea eax, [rel .data_mid]
+shl eax, 1
+
+vmovapd ymm7, [rel .data]
+
+vmovaps xmm0, [rel .index_overflow]
+vmovaps xmm1, [rel .mask_1111]
+vgatherqpd xmm7, [xmm0 * 8 + eax + 8], xmm1
+
+hlt
+
+align 32
+
+.mask_1111:
+dq 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000
+
+; Indexing is a signed 64-bit integer.
+.index_overflow:
+dq 0, 0, 0, 0
+
+; Random data, 512-byte per line
+.data:
+db 0x1d, 0x30, 0x09, 0x22, 0x56, 0xa2, 0xcd, 0xf1, 0xb7, 0x62, 0x91, 0x40, 0x67, 0x07, 0x35, 0x0f, 0x59, 0x33, 0x63, 0x52, 0x26, 0xd2, 0x2f, 0x00, 0x8a, 0x36, 0xb7, 0xf7, 0xaf, 0x4f, 0xa1, 0xc0, 0xe3, 0xbe, 0x99, 0x67, 0x5f, 0x80, 0x60, 0xcb, 0x43, 0xfa, 0x5b, 0x86, 0xb1, 0x11, 0xbc, 0xb3, 0x7b, 0x43, 0x5b, 0x45, 0x9e, 0x33, 0x89, 0xb5, 0x1b, 0xb9, 0x33, 0x4f, 0xdb, 0x5d, 0x93, 0xd6, 0x4f, 0xbc, 0x37, 0xde, 0xeb, 0xdb, 0x43, 0x2b, 0x05, 0x60, 0xb8, 0x98, 0x5c, 0xa3, 0xe3, 0x1b, 0x33, 0x03, 0x29, 0x4b, 0x12, 0x4c, 0x1e, 0xe6, 0x5e, 0x0e, 0x6c, 0xa1, 0xb9, 0x36, 0xfa, 0x6c, 0x7f, 0xc6, 0xa8, 0x38, 0x73, 0x2a, 0x0a, 0x25, 0x69, 0xa5, 0x97, 0x3f, 0x24, 0x00, 0x30, 0x4d, 0x27, 0xb3, 0x94, 0x48, 0xef, 0x47, 0x98, 0x71, 0x0d, 0x56, 0x76, 0xec, 0x41, 0x12, 0x9b, 0x7b, 0x9c, 0xf5, 0x85, 0x07, 0x2d, 0x6b, 0xc6, 0xc1, 0x2e, 0x72, 0x22, 0x5a, 0x43, 0xff, 0x1e, 0xec, 0x67, 0x2b, 0x31, 0x96, 0x14, 0x2c, 0xb1, 0x5f, 0x5d, 0x0c, 0xc9, 0xad, 0x15, 0x5f, 0xab, 0x66, 0x14, 0x1c, 0x72, 0xfa, 0x23, 0xef, 0x9f, 0x77, 0xf6, 0x50, 0xb0, 0x70, 0xb8, 0x3c, 0x85, 0x9e, 0x90, 0x69, 0x17, 0x25, 0xae, 0x6e, 0xe2, 0x16, 0x7d, 0x42, 0x38, 0xdf, 0x74, 0x72, 0x7b, 0x97, 0xa9, 0x9e, 0x40, 0x24, 0x85, 0xdc, 0x64, 0xfa, 0xb1, 0x8b, 0x95, 0xe6, 0xe4, 0x13, 0x72, 0xf1, 0x52, 0x2f, 0xa0, 0xd6, 0x52, 0xc0, 0x11, 0xa7, 0xfe, 0xd5, 0x3b, 0x56, 0xca, 0xbc, 0x01, 0xce, 0x3d, 0xd2, 0x30, 0x97, 0x1d, 0xdc, 0xeb, 0x9d, 0xa9, 0x3e, 0x09, 0xef, 0xee, 0x7f, 0x09, 0x7b, 0x82, 0x43, 0x15, 0x2e, 0xa4, 0x2e, 0x97, 0x21, 0x92, 0x7e, 0x69, 0x21, 0x25, 0xda, 0x46, 0x7c, 0x0c, 0xcd, 0x1d, 0xde, 0x42, 0x11, 0xa2, 0xef, 0xa2, 0xc8, 0x32, 0x9a, 0x82, 0xcf, 0x72, 0x7e, 0x22, 0xa6, 0x11, 0xfa, 0xec, 0x0b, 0x77, 0x99, 0x38, 0x03, 0xf6, 0x80, 0xba, 0xea, 0x75, 0x19, 0xb0, 0x48, 0x02, 0xb2, 0x6b, 0xc0, 0x8c, 0xfb, 0xfe, 0xaf, 0x94, 0x4f, 0x6f, 0xb4, 0xcb, 0x1c, 0x27, 0xf0, 0x41, 0xb6, 0x46, 0x41, 0x68, 0x3d, 0x05, 0x79, 0x6b, 0xcd, 0xb7, 0x20, 0xdc, 0x40, 0x81, 0x58, 0xcb, 0x33, 0xa3, 0xf3, 0x34, 0xdc, 0x63, 0x2d, 0xa5, 0xb5, 0xa1, 0xd1, 0xfd, 0x49, 0x5b, 0x46, 0x94, 0x01, 0xa8, 0xf2, 0xd8, 0x93, 0x2c, 0xbb, 0x57, 0xfe, 0x7c, 0x77, 0x3b, 0x19, 0x6f, 0x3c, 0xaa, 0x23, 0x5b, 0xc0, 0xe7, 0x00, 0x41, 0x97, 0x91, 0xe8, 0x00, 0x12, 0xdf, 0xf6, 0x5c, 0x2e, 0xc6, 0x8e, 0xc6, 0x77, 0x59, 0x78, 0x9b, 0xef, 0x63, 0xb0, 0xd7, 0xbb, 0xc4, 0x0b, 0x60, 0x65, 0x3f, 0xfe, 0xbf, 0x04, 0x3e, 0xae, 0xc2, 0xa5, 0x90, 0xe1, 0x2a, 0x56, 0x3f, 0x4c, 0x3f, 0x7a, 0x7d, 0xda, 0x81, 0x50, 0xea, 0x4c, 0xfe, 0xc3, 0xf8, 0x5c, 0x2b, 0x67, 0xb3, 0x9f, 0x8b, 0x95, 0xda, 0x6f, 0x5d, 0xdd, 0x82, 0x7f, 0x52, 0xa2, 0xcc, 0x57, 0xec, 0xc4, 0x14, 0xd2, 0x4f, 0x1b, 0xcb, 0xea, 0xaf, 0x0e, 0x0f, 0x53, 0xaa, 0x56, 0x63, 0xea, 0x36, 0xa6, 0x89, 0x1a, 0x66, 0xc0, 0x4e, 0xf4, 0x1e, 0x02, 0x43, 0xde, 0xde, 0xc8, 0x9e, 0x88, 0x6e, 0x32, 0xd4, 0xcb, 0x47, 0x24, 0x7c, 0x28, 0x38, 0xd4, 0x95, 0xb6, 0xa3, 0x91, 0x69, 0xc7, 0x8d, 0xfd, 0x15, 0xf5, 0xbf, 0xb1, 0x98, 0x8c, 0x57, 0x51, 0xbf, 0x83, 0x6a, 0x35, 0x10, 0x03, 0x50, 0xe5, 0xf7, 0xfa, 0xf8, 0xa5, 0xb0, 0xdb, 0xfb, 0x42, 0x93, 0xbb, 0x17, 0xf7, 0x36, 0xbe, 0x26, 0x66, 0x61, 0xe2
+
+db 0xaf, 0xe0, 0x22, 0x25, 0x2a, 0xae, 0x78, 0x67, 0x8f, 0x7e, 0x9e, 0x59, 0xd7, 0xa3, 0x71, 0xcc, 0x43, 0x85, 0x09, 0xf9, 0x18, 0x52, 0x7b, 0x01, 0x73, 0xcb, 0x31, 0x18, 0x66, 0x79, 0x67, 0x10, 0x67, 0xd8, 0xdf, 0x43, 0xaf, 0x2d, 0x9a, 0x09, 0x9c, 0xd1, 0x37, 0x7e, 0xf5, 0x1c, 0x3c, 0x4f, 0x15, 0xe1, 0x6f, 0xfd, 0x13, 0x3d, 0x53, 0x81, 0xa9, 0x93, 0x5f, 0x92, 0x41, 0x48, 0xec, 0x87, 0x87, 0x1d, 0x0b, 0xaa, 0xaa, 0xd3, 0xc2, 0x98, 0x20, 0xce, 0x28, 0xaf, 0x9d, 0x84, 0x69, 0x4a, 0xfd, 0xc0, 0x9c, 0x2e, 0x50, 0x20, 0xb2, 0x00, 0xc1, 0x81, 0x2a, 0x32, 0x8e, 0x95, 0x20, 0xa7, 0xca, 0x39, 0x28, 0x12, 0x23, 0x0e, 0x43, 0xd3, 0x82, 0x76, 0x73, 0x3c, 0xbf, 0xa9, 0x98, 0xf6, 0x39, 0x6d, 0xd9, 0x15, 0x33, 0x1e, 0x07, 0x7c, 0x08, 0x12, 0x23, 0xbd, 0xd3, 0x34, 0x2d, 0x9a, 0x23, 0x21, 0x46, 0xf3, 0x9a, 0x04, 0x25, 0x62, 0xeb, 0x7e, 0x9a, 0xaa, 0xb6, 0x26, 0xaa, 0x85, 0x01, 0x3a, 0xd8, 0xfc, 0x57, 0x98, 0xb9, 0xe4, 0xc4, 0xe9, 0x11, 0x3e, 0x22, 0x95, 0x3b, 0x41, 0x2b, 0x02, 0x04, 0x6c, 0x75, 0xa5, 0xf2, 0xaa, 0x09, 0x9e, 0x6f, 0xab, 0x1d, 0x2a, 0x5c, 0xde, 0x21, 0xb1, 0x96, 0x2d, 0x86, 0x3f, 0xd0, 0x07, 0x18, 0x1f, 0x87, 0xc2, 0x8f, 0xdf, 0x6a, 0x57, 0x6d, 0x3f, 0x80, 0xc5, 0x08, 0x19, 0xa5, 0x09, 0x65, 0x3d, 0xdc, 0x9e, 0x80, 0x3c, 0x2a, 0x0e, 0x7a, 0x40, 0x04, 0x0b, 0xcc, 0x61, 0xdb, 0x73, 0xfc, 0xa5, 0x0a, 0x42, 0x18, 0xc1, 0xd5, 0xbd, 0x18, 0x78, 0xa1, 0xe4, 0xde, 0x44, 0xec, 0x79, 0xb0, 0x27, 0xaa, 0x45, 0x21, 0x57, 0x19, 0x75, 0x09, 0x5c, 0x58, 0xd5, 0xb9, 0x6f, 0x3b, 0x48, 0x59, 0x41, 0x3e, 0xfd, 0x17, 0x43, 0x27, 0xc3, 0x8d, 0x76, 0x8e, 0x38, 0x47, 0xe7, 0xd2, 0xea, 0x54, 0x73, 0x8a, 0x65, 0x4c, 0x49, 0x91, 0xaf, 0x29, 0x65, 0x0d, 0x81, 0xa4, 0x77, 0xd7, 0x32, 0xd0, 0x69, 0xd9, 0x6b, 0xa3, 0x9b, 0x24, 0xd6, 0x0a, 0xd2, 0x77, 0x38, 0x59, 0x0b, 0xc8, 0x5c, 0xc7, 0x0b, 0x1d, 0xd1, 0xfa, 0xa7, 0x45, 0x3c, 0xeb, 0x5c, 0x8e, 0x25, 0x35, 0x81, 0x6d, 0x6d, 0xfe, 0xb4, 0x63, 0x89, 0xe4, 0xf0, 0xa8, 0xda, 0xb7, 0xd4, 0xff, 0x5d, 0x28, 0x97, 0x11, 0xf9, 0x8d, 0xab, 0x29, 0xd5, 0xd3, 0x1c, 0x70, 0x20, 0x4c, 0x41, 0x16, 0x42, 0xfd, 0xfc, 0x62, 0x82, 0x40, 0x59, 0x34, 0x28, 0xd0, 0xd5, 0xfc, 0xac, 0x97, 0xb8, 0x82, 0x0e, 0x4b, 0xae, 0x51, 0x28, 0x1a, 0xf1, 0x87, 0xd3, 0x20, 0xa3, 0xe7, 0x74, 0x69, 0x3c, 0x54, 0x8d, 0xc5, 0x56, 0x1d, 0xcd, 0x75, 0xae, 0x88, 0x17, 0x30, 0xdf, 0x46, 0x4a, 0xbc, 0x64, 0xff, 0xa2, 0xc1, 0x62, 0xbd, 0x88, 0x7b, 0x3e, 0xa1, 0x0c, 0xa9, 0x13, 0x0e, 0xc1, 0xb4, 0x24, 0xe6, 0x96, 0x1b, 0x9c, 0x9b, 0xac, 0x44, 0x33, 0x5b, 0xda, 0xd5, 0x88, 0x4d, 0xfe, 0x81, 0x09, 0x07, 0x17, 0xcf, 0x14, 0x05, 0xaf, 0xf8, 0x72, 0x14, 0x49, 0x5f, 0x06, 0x62, 0xab, 0xe0, 0x42, 0x70, 0x12, 0x59, 0x41, 0x0f, 0x18, 0x83, 0x68, 0x6d, 0xc6, 0x3c, 0xea, 0xe0, 0x6d, 0xd4, 0xae, 0xa6, 0xf1, 0x63, 0x21, 0x7f, 0xb5, 0x9d, 0x22, 0xf4, 0xd2, 0x49, 0x49, 0xed, 0x07, 0xb1, 0x11, 0xf9, 0x2e, 0x74, 0xbe, 0x35, 0x47, 0xdc, 0xef, 0x85, 0x0b, 0x4d, 0x46, 0xe6, 0x1f, 0x60, 0x6a, 0xa1, 0x8a, 0x4d, 0x46, 0x87, 0x30, 0x8e, 0x9a, 0xba, 0x97, 0x3e, 0x15, 0xb7, 0x33, 0x76, 0x81, 0x69, 0xdb, 0x82, 0x5e, 0xe6, 0x7b, 0xec, 0xd2, 0x80, 0x7f, 0x17, 0xf1, 0x73, 0xe2
+
+.data_mid:
+db 0xd7, 0xa9, 0x35, 0x61, 0x5f, 0xb8, 0xb6, 0x2d, 0x29, 0x34, 0x63, 0xbf, 0xe2, 0x1c, 0x34, 0xe9, 0xf5, 0xff, 0x34, 0x8a, 0x2f, 0xea, 0xd4, 0x3f, 0x3b, 0xfe, 0x6e, 0xdf, 0xa6, 0xd8, 0xc6, 0xb8, 0xc5, 0xff, 0x12, 0x97, 0x53, 0xda, 0x86, 0xa8, 0x0a, 0x12, 0x4e, 0x5d, 0x96, 0x65, 0x51, 0x22, 0xe2, 0x9d, 0x08, 0x71, 0x84, 0x19, 0x8b, 0xbf, 0x29, 0xd3, 0x3f, 0xab, 0xde, 0xe4, 0x27, 0x8b, 0x99, 0xcc, 0xb1, 0x7c, 0xa5, 0x71, 0x91, 0x9a, 0x0b, 0xad, 0x75, 0x86, 0xe3, 0x9c, 0x4e, 0x0c, 0x01, 0xb3, 0x12, 0x33, 0x90, 0x81, 0x7c, 0x71, 0x2c, 0x70, 0x61, 0xd5, 0x39, 0x0c, 0x45, 0xfc, 0x27, 0xaf, 0xbb, 0xd9, 0x26, 0x1b, 0x33, 0xb4, 0x0d, 0xf8, 0xd6, 0x2d, 0x09, 0xc7, 0x8c, 0xbf, 0x48, 0x53, 0x14, 0x94, 0x76, 0x25, 0xc7, 0x0c, 0x69, 0x49, 0x82, 0xb4, 0x2f, 0x48, 0x38, 0x44, 0x9d, 0x90, 0x6d, 0x66, 0x35, 0xe9, 0x3e, 0x2f, 0x2a, 0xb7, 0xe1, 0xb1, 0x2b, 0x99, 0x08, 0x6f, 0x5c, 0x6c, 0xdf, 0xdb, 0x10, 0xe2, 0xaa, 0x86, 0xe7, 0xf8, 0x9e, 0x62, 0xde, 0xa5, 0x81, 0x6b, 0x20, 0x47, 0xa9, 0x06, 0x49, 0xc0, 0x78, 0x8c, 0x70, 0x93, 0x7e, 0xda, 0xda, 0x5e, 0x3b, 0x23, 0xf9, 0xcc, 0x87, 0xdf, 0x48, 0x4f, 0xd6, 0x77, 0xce, 0x45, 0xe1, 0xdc, 0x0c, 0x7a, 0x0c, 0x50, 0x15, 0x63, 0x8c, 0x48, 0xd3, 0x8e, 0xfa, 0xcc, 0xac, 0x1a, 0x83, 0xde, 0xb1, 0x87, 0x2a, 0x58, 0x5c, 0xa5, 0x20, 0x3d, 0xaa, 0x1e, 0x5d, 0x71, 0xa6, 0x57, 0x75, 0x82, 0xb7, 0x33, 0x9e, 0x6b, 0xf3, 0x35, 0x02, 0x98, 0x03, 0xe1, 0x3b, 0xd2, 0x9f, 0x7a, 0x06, 0x85, 0xef, 0x7d, 0xd9, 0xf2, 0x0c, 0x9e, 0xce, 0xb9, 0xce, 0x13, 0x4a, 0x9e, 0x8a, 0x29, 0xe6, 0xe5, 0xe4, 0x39, 0xba, 0xfd, 0xa3, 0x33, 0xa8, 0x13, 0x9e, 0xa5, 0x11, 0x37, 0x69, 0xbc, 0xda, 0x11, 0x49, 0x2d, 0x4a, 0xef, 0x20, 0x8b, 0x7a, 0xb8, 0x9c, 0xc3, 0xaf, 0x26, 0x71, 0xd9, 0xa2, 0xf6, 0x0f, 0x85, 0x87, 0xa8, 0x6c, 0xf9, 0x99, 0xa2, 0xb2, 0x36, 0x2d, 0x78, 0x10, 0xe4, 0x33, 0x8d, 0xa4, 0x63, 0xea, 0x02, 0xb9, 0xac, 0x2f, 0x90, 0x39, 0x2d, 0x0e, 0x2e, 0xf5, 0x08, 0xa5, 0x5c, 0x8e, 0x71, 0x30, 0x0d, 0x1b, 0x84, 0x7a, 0xd7, 0xd4, 0xab, 0x81, 0x82, 0x18, 0x37, 0xf3, 0x28, 0x6f, 0x4e, 0x28, 0x71, 0xda, 0xc9, 0x99, 0x46, 0x14, 0x46, 0x77, 0x01, 0x16, 0x21, 0xae, 0x83, 0x93, 0x86, 0x7f, 0x5a, 0xee, 0xd5, 0xdf, 0x48, 0x5b, 0x15, 0xc8, 0x09, 0x30, 0x8f, 0x01, 0xcc, 0x95, 0x30, 0xd9, 0xf7, 0x72, 0x97, 0xfd, 0x9d, 0xec, 0x9f, 0xbf, 0x5c, 0xbf, 0x4f, 0xca, 0x33, 0xb4, 0xd2, 0xa2, 0xb9, 0x08, 0x9c, 0x40, 0x25, 0x3f, 0x86, 0xdc, 0x83, 0x70, 0x2f, 0xfb, 0x2a, 0xf8, 0x61, 0x1f, 0xa1, 0x1f, 0x36, 0x04, 0xe2, 0xef, 0x1c, 0xa4, 0xcd, 0x3c, 0x7f, 0xc5, 0x73, 0x9c, 0x2e, 0xeb, 0x03, 0x79, 0xd1, 0x02, 0xfc, 0x6f, 0xbd, 0x5a, 0x95, 0xb2, 0xf6, 0x25, 0x96, 0xe6, 0x80, 0x0a, 0xc5, 0xc7, 0xca, 0x8d, 0x31, 0xae, 0xf0, 0x49, 0xcf, 0x43, 0x06, 0x27, 0x7f, 0x25, 0xc7, 0x4c, 0xb7, 0xfc, 0x73, 0xd3, 0x04, 0xd3, 0xb9, 0x9f, 0x74, 0xed, 0x9e, 0x3c, 0xf0, 0xcf, 0x26, 0x2b, 0xd9, 0xcb, 0x78, 0x2a, 0xef, 0x72, 0xf7, 0xb6, 0x78, 0x30, 0x2d, 0x8c, 0x83, 0x73, 0x66, 0x74, 0x3d, 0x66, 0x0a, 0x74, 0x5a, 0x3f, 0x9f, 0x6e, 0x56, 0x68, 0x01, 0xc2, 0xca, 0x2b, 0xa1, 0x25, 0x36, 0x9c, 0x3b, 0xa4, 0x5e, 0x44, 0xf1, 0x18, 0x1d, 0xb6, 0x1a, 0x3a, 0xee, 0x8d, 0x67, 0x34, 0x9c
+
+db 0xdd, 0x48, 0x14, 0xc2, 0x5f, 0xd8, 0xe5, 0x71, 0x22, 0xbf, 0xbc, 0x84, 0xda, 0xc1, 0xb1, 0x22, 0x55, 0xa4, 0x63, 0x41, 0x77, 0xac, 0x40, 0x2d, 0x44, 0x73, 0x8c, 0x14, 0xba, 0x5e, 0x63, 0x68, 0x65, 0x61, 0x6d, 0xec, 0xe2, 0x6d, 0x37, 0x22, 0x04, 0xeb, 0xc7, 0xd4, 0xc9, 0x62, 0x56, 0x13, 0x96, 0x29, 0x03, 0xf4, 0x55, 0xe2, 0x58, 0x7d, 0xda, 0x52, 0x2e, 0x94, 0x07, 0xe6, 0xef, 0xc0, 0xee, 0x9e, 0x0b, 0xf7, 0xcd, 0x13, 0x8b, 0x7d, 0xea, 0xdc, 0xf8, 0xf1, 0xcb, 0xad, 0x49, 0x97, 0xc9, 0x98, 0x0b, 0xcf, 0x84, 0x8e, 0x8e, 0xbb, 0x06, 0x2e, 0x54, 0xf5, 0xa7, 0xbd, 0x70, 0x7e, 0x38, 0x69, 0x8d, 0xb0, 0x01, 0x7b, 0x41, 0x80, 0x09, 0x44, 0xfd, 0x7e, 0x21, 0xb4, 0xbe, 0x6b, 0x4a, 0xb7, 0xca, 0x2d, 0x19, 0xfe, 0x6d, 0xd6, 0x11, 0x29, 0xbb, 0xb2, 0x16, 0xf1, 0xe7, 0x92, 0x71, 0xda, 0x7e, 0x68, 0x3a, 0xe0, 0xea, 0x89, 0x8d, 0xe0, 0x44, 0x48, 0x25, 0x92, 0x37, 0x54, 0x26, 0xf2, 0xab, 0xb3, 0x3b, 0xdb, 0xbb, 0x2b, 0x5c, 0xf5, 0xbc, 0xc7, 0x97, 0xdb, 0xc7, 0x49, 0x25, 0x7c, 0xc2, 0x80, 0x02, 0x69, 0xd4, 0xda, 0xda, 0xe1, 0x04, 0xf3, 0x19, 0xb8, 0xc9, 0xb2, 0xfb, 0x1e, 0x47, 0xa9, 0x0c, 0xa3, 0x48, 0xce, 0xc2, 0x9e, 0x3b, 0x28, 0x23, 0x5a, 0x20, 0x44, 0x77, 0x40, 0xe2, 0xd7, 0x20, 0xd5, 0x71, 0x6f, 0xd4, 0x3c, 0x68, 0x38, 0x9b, 0x89, 0x2e, 0x2d, 0xa8, 0x1f, 0x99, 0xb5, 0x8a, 0x66, 0x07, 0x59, 0x75, 0x9e, 0xf8, 0xd9, 0xbe, 0x85, 0x6a, 0x20, 0x92, 0x9d, 0xd2, 0x5e, 0x45, 0xc0, 0x60, 0xbe, 0x85, 0x0b, 0x84, 0x47, 0xf5, 0xa8, 0x43, 0x87, 0xf1, 0x21, 0x21, 0xb0, 0x3b, 0x04, 0x13, 0x16, 0x3e, 0xdf, 0xc3, 0xc6, 0x04, 0x73, 0xcd, 0x92, 0x76, 0xfb, 0xe7, 0x9c, 0xd3, 0x46, 0x11, 0x78, 0xca, 0x12, 0xd9, 0x4a, 0x35, 0xf1, 0x6e, 0x89, 0x8b, 0xe9, 0x7a, 0x04, 0xba, 0x18, 0x25, 0x7c, 0x9e, 0xe6, 0x4f, 0xc2, 0x56, 0x05, 0x72, 0xc3, 0x76, 0xee, 0x7d, 0x77, 0x19, 0x7a, 0x73, 0x2c, 0x81, 0xb8, 0xc7, 0xd9, 0x7f, 0x17, 0x5d, 0x30, 0xda, 0x77, 0x3c, 0x14, 0x88, 0xe8, 0xe4, 0xbf, 0xee, 0x21, 0x1c, 0x29, 0x4e, 0x58, 0xa8, 0x8a, 0x5c, 0xae, 0xa2, 0x1c, 0x7c, 0x25, 0x7c, 0x1c, 0x39, 0xa4, 0x28, 0x4b, 0x78, 0x52, 0xae, 0x2c, 0xbb, 0x5f, 0xbf, 0x51, 0x09, 0x20, 0x76, 0xb2, 0x7d, 0xb1, 0x63, 0x84, 0xc5, 0x49, 0x8a, 0x73, 0xdb, 0x76, 0x1d, 0x25, 0x31, 0xf2, 0x1e, 0x19, 0x38, 0xc8, 0x3b, 0x51, 0x3c, 0x13, 0x52, 0x84, 0xae, 0xc2, 0xe4, 0x8a, 0x57, 0x0d, 0xde, 0x8d, 0x18, 0x48, 0x9a, 0xbd, 0xbf, 0xf3, 0xea, 0x79, 0x17, 0x06, 0x96, 0x72, 0x08, 0x60, 0x95, 0xf9, 0x6f, 0x25, 0x0c, 0xb7, 0x9d, 0x98, 0x23, 0x01, 0xc8, 0x7a, 0xdb, 0x75, 0x63, 0x64, 0x14, 0x5e, 0x10, 0xf5, 0x16, 0x48, 0xbc, 0xc6, 0x7e, 0x24, 0xf3, 0xad, 0x57, 0x3f, 0x7d, 0x6c, 0xab, 0x18, 0x8c, 0x12, 0xc5, 0x0c, 0xd8, 0xb5, 0x1e, 0x43, 0x7c, 0x23, 0x17, 0x48, 0xba, 0x76, 0x3b, 0xd9, 0x2b, 0xae, 0x1b, 0xef, 0x58, 0xfa, 0x87, 0xad, 0x9b, 0x6d, 0xf9, 0xab, 0xa8, 0x3c, 0xfc, 0x59, 0x67, 0xa6, 0x2c, 0xc7, 0x75, 0xa4, 0x97, 0xca, 0x18, 0x18, 0x04, 0x2c, 0xb3, 0x0e, 0xa9, 0x69, 0x33, 0x67, 0xa2, 0xc6, 0xbc, 0x98, 0x48, 0x71, 0x11, 0x05, 0x30, 0xf6, 0xa9, 0x61, 0x40, 0x46, 0xf1, 0x41, 0x37, 0xd0, 0x6b, 0x7c, 0x1f, 0x03, 0x5c, 0xe9, 0xf4, 0x59, 0x1d, 0x35, 0xf0, 0x98, 0x42, 0x4a, 0x92, 0x2a, 0xc3, 0x9a, 0xb8, 0xa5

unittests/32Bit_ASM/VEX/vgather_qpd_256bit_1xdisp_overflow.asm

@@ -0,0 +1,69 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "XMM1":  ["0", "0", "0", "0"],
+    "XMM7": ["0x2db6b85f6135a9d7", "0xe273f1177f80d2ec", "0x7be65e82db698176", "0x3fd4ea2f8a34fff5"]
+  },
+  "Mode": "32BIT",
+  "Env": { "FEX_HOSTFEATURES" : "enableavx" },
+  "HostFeatures": ["AVX"]
+}
+%endif
+
+; 256-bit
+; 1x displacement
+; 64-bit indexes
+
+; Calculate an address that heavily overflows
+lea eax, [rel .data_mid]
+
+mov ebx, -1
+sub ebx, eax
+mov [rel .index_overflow + 0 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 8
+mov [rel .index_overflow + 1 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 16
+mov [rel .index_overflow + 2 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+add bx, 16
+mov [rel .index_overflow + 3 * 8], ebx
+
+; Calculate new base which offsets from the overflow
+lea eax, [rel .data_mid]
+shl eax, 1
+
+vmovapd ymm7, [rel .data]
+
+vmovaps ymm0, [rel .index_overflow]
+vmovaps ymm1, [rel .mask_1111]
+vgatherqpd ymm7, [ymm0 * 1 + eax + 1], ymm1
+
+hlt
+
+align 32
+
+.mask_1111:
+dq 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000
+
+; Indexing is a signed 64-bit integer.
+.index_overflow:
+dq 0, 0, 0, 0
+
+; Random data, 512-byte per line
+.data:
+db 0x1d, 0x30, 0x09, 0x22, 0x56, 0xa2, 0xcd, 0xf1, 0xb7, 0x62, 0x91, 0x40, 0x67, 0x07, 0x35, 0x0f, 0x59, 0x33, 0x63, 0x52, 0x26, 0xd2, 0x2f, 0x00, 0x8a, 0x36, 0xb7, 0xf7, 0xaf, 0x4f, 0xa1, 0xc0, 0xe3, 0xbe, 0x99, 0x67, 0x5f, 0x80, 0x60, 0xcb, 0x43, 0xfa, 0x5b, 0x86, 0xb1, 0x11, 0xbc, 0xb3, 0x7b, 0x43, 0x5b, 0x45, 0x9e, 0x33, 0x89, 0xb5, 0x1b, 0xb9, 0x33, 0x4f, 0xdb, 0x5d, 0x93, 0xd6, 0x4f, 0xbc, 0x37, 0xde, 0xeb, 0xdb, 0x43, 0x2b, 0x05, 0x60, 0xb8, 0x98, 0x5c, 0xa3, 0xe3, 0x1b, 0x33, 0x03, 0x29, 0x4b, 0x12, 0x4c, 0x1e, 0xe6, 0x5e, 0x0e, 0x6c, 0xa1, 0xb9, 0x36, 0xfa, 0x6c, 0x7f, 0xc6, 0xa8, 0x38, 0x73, 0x2a, 0x0a, 0x25, 0x69, 0xa5, 0x97, 0x3f, 0x24, 0x00, 0x30, 0x4d, 0x27, 0xb3, 0x94, 0x48, 0xef, 0x47, 0x98, 0x71, 0x0d, 0x56, 0x76, 0xec, 0x41, 0x12, 0x9b, 0x7b, 0x9c, 0xf5, 0x85, 0x07, 0x2d, 0x6b, 0xc6, 0xc1, 0x2e, 0x72, 0x22, 0x5a, 0x43, 0xff, 0x1e, 0xec, 0x67, 0x2b, 0x31, 0x96, 0x14, 0x2c, 0xb1, 0x5f, 0x5d, 0x0c, 0xc9, 0xad, 0x15, 0x5f, 0xab, 0x66, 0x14, 0x1c, 0x72, 0xfa, 0x23, 0xef, 0x9f, 0x77, 0xf6, 0x50, 0xb0, 0x70, 0xb8, 0x3c, 0x85, 0x9e, 0x90, 0x69, 0x17, 0x25, 0xae, 0x6e, 0xe2, 0x16, 0x7d, 0x42, 0x38, 0xdf, 0x74, 0x72, 0x7b, 0x97, 0xa9, 0x9e, 0x40, 0x24, 0x85, 0xdc, 0x64, 0xfa, 0xb1, 0x8b, 0x95, 0xe6, 0xe4, 0x13, 0x72, 0xf1, 0x52, 0x2f, 0xa0, 0xd6, 0x52, 0xc0, 0x11, 0xa7, 0xfe, 0xd5, 0x3b, 0x56, 0xca, 0xbc, 0x01, 0xce, 0x3d, 0xd2, 0x30, 0x97, 0x1d, 0xdc, 0xeb, 0x9d, 0xa9, 0x3e, 0x09, 0xef, 0xee, 0x7f, 0x09, 0x7b, 0x82, 0x43, 0x15, 0x2e, 0xa4, 0x2e, 0x97, 0x21, 0x92, 0x7e, 0x69, 0x21, 0x25, 0xda, 0x46, 0x7c, 0x0c, 0xcd, 0x1d, 0xde, 0x42, 0x11, 0xa2, 0xef, 0xa2, 0xc8, 0x32, 0x9a, 0x82, 0xcf, 0x72, 0x7e, 0x22, 0xa6, 0x11, 0xfa, 0xec, 0x0b, 0x77, 0x99, 0x38, 0x03, 0xf6, 0x80, 0xba, 0xea, 0x75, 0x19, 0xb0, 0x48, 0x02, 0xb2, 0x6b, 0xc0, 0x8c, 0xfb, 0xfe, 0xaf, 0x94, 0x4f, 0x6f, 0xb4, 0xcb, 0x1c, 0x27, 0xf0, 0x41, 0xb6, 0x46, 0x41, 0x68, 0x3d, 0x05, 0x79, 0x6b, 0xcd, 0xb7, 0x20, 0xdc, 0x40, 0x81, 0x58, 0xcb, 0x33, 0xa3, 0xf3, 0x34, 0xdc, 0x63, 0x2d, 0xa5, 0xb5, 0xa1, 0xd1, 0xfd, 0x49, 0x5b, 0x46, 0x94, 0x01, 0xa8, 0xf2, 0xd8, 0x93, 0x2c, 0xbb, 0x57, 0xfe, 0x7c, 0x77, 0x3b, 0x19, 0x6f, 0x3c, 0xaa, 0x23, 0x5b, 0xc0, 0xe7, 0x00, 0x41, 0x97, 0x91, 0xe8, 0x00, 0x12, 0xdf, 0xf6, 0x5c, 0x2e, 0xc6, 0x8e, 0xc6, 0x77, 0x59, 0x78, 0x9b, 0xef, 0x63, 0xb0, 0xd7, 0xbb, 0xc4, 0x0b, 0x60, 0x65, 0x3f, 0xfe, 0xbf, 0x04, 0x3e, 0xae, 0xc2, 0xa5, 0x90, 0xe1, 0x2a, 0x56, 0x3f, 0x4c, 0x3f, 0x7a, 0x7d, 0xda, 0x81, 0x50, 0xea, 0x4c, 0xfe, 0xc3, 0xf8, 0x5c, 0x2b, 0x67, 0xb3, 0x9f, 0x8b, 0x95, 0xda, 0x6f, 0x5d, 0xdd, 0x82, 0x7f, 0x52, 0xa2, 0xcc, 0x57, 0xec, 0xc4, 0x14, 0xd2, 0x4f, 0x1b, 0xcb, 0xea, 0xaf, 0x0e, 0x0f, 0x53, 0xaa, 0x56, 0x63, 0xea, 0x36, 0xa6, 0x89, 0x1a, 0x66, 0xc0, 0x4e, 0xf4, 0x1e, 0x02, 0x43, 0xde, 0xde, 0xc8, 0x9e, 0x88, 0x6e, 0x32, 0xd4, 0xcb, 0x47, 0x24, 0x7c, 0x28, 0x38, 0xd4, 0x95, 0xb6, 0xa3, 0x91, 0x69, 0xc7, 0x8d, 0xfd, 0x15, 0xf5, 0xbf, 0xb1, 0x98, 0x8c, 0x57, 0x51, 0xbf, 0x83, 0x6a, 0x35, 0x10, 0x03, 0x50, 0xe5, 0xf7, 0xfa, 0xf8, 0xa5, 0xb0, 0xdb, 0xfb, 0x42, 0x93, 0xbb, 0x17, 0xf7, 0x36, 0xbe, 0x26, 0x66, 0x61, 0xe2
+
+db 0xaf, 0xe0, 0x22, 0x25, 0x2a, 0xae, 0x78, 0x67, 0x8f, 0x7e, 0x9e, 0x59, 0xd7, 0xa3, 0x71, 0xcc, 0x43, 0x85, 0x09, 0xf9, 0x18, 0x52, 0x7b, 0x01, 0x73, 0xcb, 0x31, 0x18, 0x66, 0x79, 0x67, 0x10, 0x67, 0xd8, 0xdf, 0x43, 0xaf, 0x2d, 0x9a, 0x09, 0x9c, 0xd1, 0x37, 0x7e, 0xf5, 0x1c, 0x3c, 0x4f, 0x15, 0xe1, 0x6f, 0xfd, 0x13, 0x3d, 0x53, 0x81, 0xa9, 0x93, 0x5f, 0x92, 0x41, 0x48, 0xec, 0x87, 0x87, 0x1d, 0x0b, 0xaa, 0xaa, 0xd3, 0xc2, 0x98, 0x20, 0xce, 0x28, 0xaf, 0x9d, 0x84, 0x69, 0x4a, 0xfd, 0xc0, 0x9c, 0x2e, 0x50, 0x20, 0xb2, 0x00, 0xc1, 0x81, 0x2a, 0x32, 0x8e, 0x95, 0x20, 0xa7, 0xca, 0x39, 0x28, 0x12, 0x23, 0x0e, 0x43, 0xd3, 0x82, 0x76, 0x73, 0x3c, 0xbf, 0xa9, 0x98, 0xf6, 0x39, 0x6d, 0xd9, 0x15, 0x33, 0x1e, 0x07, 0x7c, 0x08, 0x12, 0x23, 0xbd, 0xd3, 0x34, 0x2d, 0x9a, 0x23, 0x21, 0x46, 0xf3, 0x9a, 0x04, 0x25, 0x62, 0xeb, 0x7e, 0x9a, 0xaa, 0xb6, 0x26, 0xaa, 0x85, 0x01, 0x3a, 0xd8, 0xfc, 0x57, 0x98, 0xb9, 0xe4, 0xc4, 0xe9, 0x11, 0x3e, 0x22, 0x95, 0x3b, 0x41, 0x2b, 0x02, 0x04, 0x6c, 0x75, 0xa5, 0xf2, 0xaa, 0x09, 0x9e, 0x6f, 0xab, 0x1d, 0x2a, 0x5c, 0xde, 0x21, 0xb1, 0x96, 0x2d, 0x86, 0x3f, 0xd0, 0x07, 0x18, 0x1f, 0x87, 0xc2, 0x8f, 0xdf, 0x6a, 0x57, 0x6d, 0x3f, 0x80, 0xc5, 0x08, 0x19, 0xa5, 0x09, 0x65, 0x3d, 0xdc, 0x9e, 0x80, 0x3c, 0x2a, 0x0e, 0x7a, 0x40, 0x04, 0x0b, 0xcc, 0x61, 0xdb, 0x73, 0xfc, 0xa5, 0x0a, 0x42, 0x18, 0xc1, 0xd5, 0xbd, 0x18, 0x78, 0xa1, 0xe4, 0xde, 0x44, 0xec, 0x79, 0xb0, 0x27, 0xaa, 0x45, 0x21, 0x57, 0x19, 0x75, 0x09, 0x5c, 0x58, 0xd5, 0xb9, 0x6f, 0x3b, 0x48, 0x59, 0x41, 0x3e, 0xfd, 0x17, 0x43, 0x27, 0xc3, 0x8d, 0x76, 0x8e, 0x38, 0x47, 0xe7, 0xd2, 0xea, 0x54, 0x73, 0x8a, 0x65, 0x4c, 0x49, 0x91, 0xaf, 0x29, 0x65, 0x0d, 0x81, 0xa4, 0x77, 0xd7, 0x32, 0xd0, 0x69, 0xd9, 0x6b, 0xa3, 0x9b, 0x24, 0xd6, 0x0a, 0xd2, 0x77, 0x38, 0x59, 0x0b, 0xc8, 0x5c, 0xc7, 0x0b, 0x1d, 0xd1, 0xfa, 0xa7, 0x45, 0x3c, 0xeb, 0x5c, 0x8e, 0x25, 0x35, 0x81, 0x6d, 0x6d, 0xfe, 0xb4, 0x63, 0x89, 0xe4, 0xf0, 0xa8, 0xda, 0xb7, 0xd4, 0xff, 0x5d, 0x28, 0x97, 0x11, 0xf9, 0x8d, 0xab, 0x29, 0xd5, 0xd3, 0x1c, 0x70, 0x20, 0x4c, 0x41, 0x16, 0x42, 0xfd, 0xfc, 0x62, 0x82, 0x40, 0x59, 0x34, 0x28, 0xd0, 0xd5, 0xfc, 0xac, 0x97, 0xb8, 0x82, 0x0e, 0x4b, 0xae, 0x51, 0x28, 0x1a, 0xf1, 0x87, 0xd3, 0x20, 0xa3, 0xe7, 0x74, 0x69, 0x3c, 0x54, 0x8d, 0xc5, 0x56, 0x1d, 0xcd, 0x75, 0xae, 0x88, 0x17, 0x30, 0xdf, 0x46, 0x4a, 0xbc, 0x64, 0xff, 0xa2, 0xc1, 0x62, 0xbd, 0x88, 0x7b, 0x3e, 0xa1, 0x0c, 0xa9, 0x13, 0x0e, 0xc1, 0xb4, 0x24, 0xe6, 0x96, 0x1b, 0x9c, 0x9b, 0xac, 0x44, 0x33, 0x5b, 0xda, 0xd5, 0x88, 0x4d, 0xfe, 0x81, 0x09, 0x07, 0x17, 0xcf, 0x14, 0x05, 0xaf, 0xf8, 0x72, 0x14, 0x49, 0x5f, 0x06, 0x62, 0xab, 0xe0, 0x42, 0x70, 0x12, 0x59, 0x41, 0x0f, 0x18, 0x83, 0x68, 0x6d, 0xc6, 0x3c, 0xea, 0xe0, 0x6d, 0xd4, 0xae, 0xa6, 0xf1, 0x63, 0x21, 0x7f, 0xb5, 0x9d, 0x22, 0xf4, 0xd2, 0x49, 0x49, 0xed, 0x07, 0xb1, 0x11, 0xf9, 0x2e, 0x74, 0xbe, 0x35, 0x47, 0xdc, 0xef, 0x85, 0x0b, 0x4d, 0x46, 0xe6, 0x1f, 0x60, 0x6a, 0xa1, 0x8a, 0x4d, 0x46, 0x87, 0x30, 0x8e, 0x9a, 0xba, 0x97, 0x3e, 0x15, 0xb7, 0x33, 0x76, 0x81, 0x69, 0xdb, 0x82, 0x5e, 0xe6, 0x7b, 0xec, 0xd2, 0x80, 0x7f, 0x17, 0xf1, 0x73, 0xe2
+
+.data_mid:
+db 0xd7, 0xa9, 0x35, 0x61, 0x5f, 0xb8, 0xb6, 0x2d, 0x29, 0x34, 0x63, 0xbf, 0xe2, 0x1c, 0x34, 0xe9, 0xf5, 0xff, 0x34, 0x8a, 0x2f, 0xea, 0xd4, 0x3f, 0x3b, 0xfe, 0x6e, 0xdf, 0xa6, 0xd8, 0xc6, 0xb8, 0xc5, 0xff, 0x12, 0x97, 0x53, 0xda, 0x86, 0xa8, 0x0a, 0x12, 0x4e, 0x5d, 0x96, 0x65, 0x51, 0x22, 0xe2, 0x9d, 0x08, 0x71, 0x84, 0x19, 0x8b, 0xbf, 0x29, 0xd3, 0x3f, 0xab, 0xde, 0xe4, 0x27, 0x8b, 0x99, 0xcc, 0xb1, 0x7c, 0xa5, 0x71, 0x91, 0x9a, 0x0b, 0xad, 0x75, 0x86, 0xe3, 0x9c, 0x4e, 0x0c, 0x01, 0xb3, 0x12, 0x33, 0x90, 0x81, 0x7c, 0x71, 0x2c, 0x70, 0x61, 0xd5, 0x39, 0x0c, 0x45, 0xfc, 0x27, 0xaf, 0xbb, 0xd9, 0x26, 0x1b, 0x33, 0xb4, 0x0d, 0xf8, 0xd6, 0x2d, 0x09, 0xc7, 0x8c, 0xbf, 0x48, 0x53, 0x14, 0x94, 0x76, 0x25, 0xc7, 0x0c, 0x69, 0x49, 0x82, 0xb4, 0x2f, 0x48, 0x38, 0x44, 0x9d, 0x90, 0x6d, 0x66, 0x35, 0xe9, 0x3e, 0x2f, 0x2a, 0xb7, 0xe1, 0xb1, 0x2b, 0x99, 0x08, 0x6f, 0x5c, 0x6c, 0xdf, 0xdb, 0x10, 0xe2, 0xaa, 0x86, 0xe7, 0xf8, 0x9e, 0x62, 0xde, 0xa5, 0x81, 0x6b, 0x20, 0x47, 0xa9, 0x06, 0x49, 0xc0, 0x78, 0x8c, 0x70, 0x93, 0x7e, 0xda, 0xda, 0x5e, 0x3b, 0x23, 0xf9, 0xcc, 0x87, 0xdf, 0x48, 0x4f, 0xd6, 0x77, 0xce, 0x45, 0xe1, 0xdc, 0x0c, 0x7a, 0x0c, 0x50, 0x15, 0x63, 0x8c, 0x48, 0xd3, 0x8e, 0xfa, 0xcc, 0xac, 0x1a, 0x83, 0xde, 0xb1, 0x87, 0x2a, 0x58, 0x5c, 0xa5, 0x20, 0x3d, 0xaa, 0x1e, 0x5d, 0x71, 0xa6, 0x57, 0x75, 0x82, 0xb7, 0x33, 0x9e, 0x6b, 0xf3, 0x35, 0x02, 0x98, 0x03, 0xe1, 0x3b, 0xd2, 0x9f, 0x7a, 0x06, 0x85, 0xef, 0x7d, 0xd9, 0xf2, 0x0c, 0x9e, 0xce, 0xb9, 0xce, 0x13, 0x4a, 0x9e, 0x8a, 0x29, 0xe6, 0xe5, 0xe4, 0x39, 0xba, 0xfd, 0xa3, 0x33, 0xa8, 0x13, 0x9e, 0xa5, 0x11, 0x37, 0x69, 0xbc, 0xda, 0x11, 0x49, 0x2d, 0x4a, 0xef, 0x20, 0x8b, 0x7a, 0xb8, 0x9c, 0xc3, 0xaf, 0x26, 0x71, 0xd9, 0xa2, 0xf6, 0x0f, 0x85, 0x87, 0xa8, 0x6c, 0xf9, 0x99, 0xa2, 0xb2, 0x36, 0x2d, 0x78, 0x10, 0xe4, 0x33, 0x8d, 0xa4, 0x63, 0xea, 0x02, 0xb9, 0xac, 0x2f, 0x90, 0x39, 0x2d, 0x0e, 0x2e, 0xf5, 0x08, 0xa5, 0x5c, 0x8e, 0x71, 0x30, 0x0d, 0x1b, 0x84, 0x7a, 0xd7, 0xd4, 0xab, 0x81, 0x82, 0x18, 0x37, 0xf3, 0x28, 0x6f, 0x4e, 0x28, 0x71, 0xda, 0xc9, 0x99, 0x46, 0x14, 0x46, 0x77, 0x01, 0x16, 0x21, 0xae, 0x83, 0x93, 0x86, 0x7f, 0x5a, 0xee, 0xd5, 0xdf, 0x48, 0x5b, 0x15, 0xc8, 0x09, 0x30, 0x8f, 0x01, 0xcc, 0x95, 0x30, 0xd9, 0xf7, 0x72, 0x97, 0xfd, 0x9d, 0xec, 0x9f, 0xbf, 0x5c, 0xbf, 0x4f, 0xca, 0x33, 0xb4, 0xd2, 0xa2, 0xb9, 0x08, 0x9c, 0x40, 0x25, 0x3f, 0x86, 0xdc, 0x83, 0x70, 0x2f, 0xfb, 0x2a, 0xf8, 0x61, 0x1f, 0xa1, 0x1f, 0x36, 0x04, 0xe2, 0xef, 0x1c, 0xa4, 0xcd, 0x3c, 0x7f, 0xc5, 0x73, 0x9c, 0x2e, 0xeb, 0x03, 0x79, 0xd1, 0x02, 0xfc, 0x6f, 0xbd, 0x5a, 0x95, 0xb2, 0xf6, 0x25, 0x96, 0xe6, 0x80, 0x0a, 0xc5, 0xc7, 0xca, 0x8d, 0x31, 0xae, 0xf0, 0x49, 0xcf, 0x43, 0x06, 0x27, 0x7f, 0x25, 0xc7, 0x4c, 0xb7, 0xfc, 0x73, 0xd3, 0x04, 0xd3, 0xb9, 0x9f, 0x74, 0xed, 0x9e, 0x3c, 0xf0, 0xcf, 0x26, 0x2b, 0xd9, 0xcb, 0x78, 0x2a, 0xef, 0x72, 0xf7, 0xb6, 0x78, 0x30, 0x2d, 0x8c, 0x83, 0x73, 0x66, 0x74, 0x3d, 0x66, 0x0a, 0x74, 0x5a, 0x3f, 0x9f, 0x6e, 0x56, 0x68, 0x01, 0xc2, 0xca, 0x2b, 0xa1, 0x25, 0x36, 0x9c, 0x3b, 0xa4, 0x5e, 0x44, 0xf1, 0x18, 0x1d, 0xb6, 0x1a, 0x3a, 0xee, 0x8d, 0x67, 0x34, 0x9c
+
+db 0xdd, 0x48, 0x14, 0xc2, 0x5f, 0xd8, 0xe5, 0x71, 0x22, 0xbf, 0xbc, 0x84, 0xda, 0xc1, 0xb1, 0x22, 0x55, 0xa4, 0x63, 0x41, 0x77, 0xac, 0x40, 0x2d, 0x44, 0x73, 0x8c, 0x14, 0xba, 0x5e, 0x63, 0x68, 0x65, 0x61, 0x6d, 0xec, 0xe2, 0x6d, 0x37, 0x22, 0x04, 0xeb, 0xc7, 0xd4, 0xc9, 0x62, 0x56, 0x13, 0x96, 0x29, 0x03, 0xf4, 0x55, 0xe2, 0x58, 0x7d, 0xda, 0x52, 0x2e, 0x94, 0x07, 0xe6, 0xef, 0xc0, 0xee, 0x9e, 0x0b, 0xf7, 0xcd, 0x13, 0x8b, 0x7d, 0xea, 0xdc, 0xf8, 0xf1, 0xcb, 0xad, 0x49, 0x97, 0xc9, 0x98, 0x0b, 0xcf, 0x84, 0x8e, 0x8e, 0xbb, 0x06, 0x2e, 0x54, 0xf5, 0xa7, 0xbd, 0x70, 0x7e, 0x38, 0x69, 0x8d, 0xb0, 0x01, 0x7b, 0x41, 0x80, 0x09, 0x44, 0xfd, 0x7e, 0x21, 0xb4, 0xbe, 0x6b, 0x4a, 0xb7, 0xca, 0x2d, 0x19, 0xfe, 0x6d, 0xd6, 0x11, 0x29, 0xbb, 0xb2, 0x16, 0xf1, 0xe7, 0x92, 0x71, 0xda, 0x7e, 0x68, 0x3a, 0xe0, 0xea, 0x89, 0x8d, 0xe0, 0x44, 0x48, 0x25, 0x92, 0x37, 0x54, 0x26, 0xf2, 0xab, 0xb3, 0x3b, 0xdb, 0xbb, 0x2b, 0x5c, 0xf5, 0xbc, 0xc7, 0x97, 0xdb, 0xc7, 0x49, 0x25, 0x7c, 0xc2, 0x80, 0x02, 0x69, 0xd4, 0xda, 0xda, 0xe1, 0x04, 0xf3, 0x19, 0xb8, 0xc9, 0xb2, 0xfb, 0x1e, 0x47, 0xa9, 0x0c, 0xa3, 0x48, 0xce, 0xc2, 0x9e, 0x3b, 0x28, 0x23, 0x5a, 0x20, 0x44, 0x77, 0x40, 0xe2, 0xd7, 0x20, 0xd5, 0x71, 0x6f, 0xd4, 0x3c, 0x68, 0x38, 0x9b, 0x89, 0x2e, 0x2d, 0xa8, 0x1f, 0x99, 0xb5, 0x8a, 0x66, 0x07, 0x59, 0x75, 0x9e, 0xf8, 0xd9, 0xbe, 0x85, 0x6a, 0x20, 0x92, 0x9d, 0xd2, 0x5e, 0x45, 0xc0, 0x60, 0xbe, 0x85, 0x0b, 0x84, 0x47, 0xf5, 0xa8, 0x43, 0x87, 0xf1, 0x21, 0x21, 0xb0, 0x3b, 0x04, 0x13, 0x16, 0x3e, 0xdf, 0xc3, 0xc6, 0x04, 0x73, 0xcd, 0x92, 0x76, 0xfb, 0xe7, 0x9c, 0xd3, 0x46, 0x11, 0x78, 0xca, 0x12, 0xd9, 0x4a, 0x35, 0xf1, 0x6e, 0x89, 0x8b, 0xe9, 0x7a, 0x04, 0xba, 0x18, 0x25, 0x7c, 0x9e, 0xe6, 0x4f, 0xc2, 0x56, 0x05, 0x72, 0xc3, 0x76, 0xee, 0x7d, 0x77, 0x19, 0x7a, 0x73, 0x2c, 0x81, 0xb8, 0xc7, 0xd9, 0x7f, 0x17, 0x5d, 0x30, 0xda, 0x77, 0x3c, 0x14, 0x88, 0xe8, 0xe4, 0xbf, 0xee, 0x21, 0x1c, 0x29, 0x4e, 0x58, 0xa8, 0x8a, 0x5c, 0xae, 0xa2, 0x1c, 0x7c, 0x25, 0x7c, 0x1c, 0x39, 0xa4, 0x28, 0x4b, 0x78, 0x52, 0xae, 0x2c, 0xbb, 0x5f, 0xbf, 0x51, 0x09, 0x20, 0x76, 0xb2, 0x7d, 0xb1, 0x63, 0x84, 0xc5, 0x49, 0x8a, 0x73, 0xdb, 0x76, 0x1d, 0x25, 0x31, 0xf2, 0x1e, 0x19, 0x38, 0xc8, 0x3b, 0x51, 0x3c, 0x13, 0x52, 0x84, 0xae, 0xc2, 0xe4, 0x8a, 0x57, 0x0d, 0xde, 0x8d, 0x18, 0x48, 0x9a, 0xbd, 0xbf, 0xf3, 0xea, 0x79, 0x17, 0x06, 0x96, 0x72, 0x08, 0x60, 0x95, 0xf9, 0x6f, 0x25, 0x0c, 0xb7, 0x9d, 0x98, 0x23, 0x01, 0xc8, 0x7a, 0xdb, 0x75, 0x63, 0x64, 0x14, 0x5e, 0x10, 0xf5, 0x16, 0x48, 0xbc, 0xc6, 0x7e, 0x24, 0xf3, 0xad, 0x57, 0x3f, 0x7d, 0x6c, 0xab, 0x18, 0x8c, 0x12, 0xc5, 0x0c, 0xd8, 0xb5, 0x1e, 0x43, 0x7c, 0x23, 0x17, 0x48, 0xba, 0x76, 0x3b, 0xd9, 0x2b, 0xae, 0x1b, 0xef, 0x58, 0xfa, 0x87, 0xad, 0x9b, 0x6d, 0xf9, 0xab, 0xa8, 0x3c, 0xfc, 0x59, 0x67, 0xa6, 0x2c, 0xc7, 0x75, 0xa4, 0x97, 0xca, 0x18, 0x18, 0x04, 0x2c, 0xb3, 0x0e, 0xa9, 0x69, 0x33, 0x67, 0xa2, 0xc6, 0xbc, 0x98, 0x48, 0x71, 0x11, 0x05, 0x30, 0xf6, 0xa9, 0x61, 0x40, 0x46, 0xf1, 0x41, 0x37, 0xd0, 0x6b, 0x7c, 0x1f, 0x03, 0x5c, 0xe9, 0xf4, 0x59, 0x1d, 0x35, 0xf0, 0x98, 0x42, 0x4a, 0x92, 0x2a, 0xc3, 0x9a, 0xb8, 0xa5

unittests/32Bit_ASM/VEX/vgather_qpd_256bit_2xdisp_overflow.asm

@@ -0,0 +1,73 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "XMM1":  ["0", "0", "0", "0"],
+    "XMM7": ["0x2db6b85f6135a9d7", "0xe273f1177f80d2ec", "0x7be65e82db698176", "0x3fd4ea2f8a34fff5"]
+  },
+  "Mode": "32BIT",
+  "Env": { "FEX_HOSTFEATURES" : "enableavx" },
+  "HostFeatures": ["AVX"]
+}
+%endif
+
+; 256-bit
+; 2x displacement
+; 64-bit indexes
+
+; Calculate an address that heavily overflows
+lea eax, [rel .data_mid]
+
+mov ebx, -1
+sub ebx, eax
+sar ebx, 1
+mov [rel .index_overflow + 0 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 8
+sar ebx, 1
+mov [rel .index_overflow + 1 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+sub ebx, 16
+sar ebx, 1
+mov [rel .index_overflow + 2 * 8], ebx
+
+mov ebx, -1
+sub ebx, eax
+add bx, 16
+sar ebx, 1
+mov [rel .index_overflow + 3 * 8], ebx
+
+; Calculate new base which offsets from the overflow
+lea eax, [rel .data_mid]
+shl eax, 1
+
+vmovapd ymm7, [rel .data]
+
+vmovaps ymm0, [rel .index_overflow]
+vmovaps ymm1, [rel .mask_1111]
+vgatherqpd ymm7, [ymm0 * 2 + eax + 2], ymm1
+
+hlt
+
+align 32
+
+.mask_1111:
+dq 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000, 0x8000_0000_0000_0000
+
+; Indexing is a signed 64-bit integer.
+.index_overflow:
+dq 0, 0, 0, 0
+
+; Random data, 512-byte per line
+.data:
+db 0x1d, 0x30, 0x09, 0x22, 0x56, 0xa2, 0xcd, 0xf1, 0xb7, 0x62, 0x91, 0x40, 0x67, 0x07, 0x35, 0x0f, 0x59, 0x33, 0x63, 0x52, 0x26, 0xd2, 0x2f, 0x00, 0x8a, 0x36, 0xb7, 0xf7, 0xaf, 0x4f, 0xa1, 0xc0, 0xe3, 0xbe, 0x99, 0x67, 0x5f, 0x80, 0x60, 0xcb, 0x43, 0xfa, 0x5b, 0x86, 0xb1, 0x11, 0xbc, 0xb3, 0x7b, 0x43, 0x5b, 0x45, 0x9e, 0x33, 0x89, 0xb5, 0x1b, 0xb9, 0x33, 0x4f, 0xdb, 0x5d, 0x93, 0xd6, 0x4f, 0xbc, 0x37, 0xde, 0xeb, 0xdb, 0x43, 0x2b, 0x05, 0x60, 0xb8, 0x98, 0x5c, 0xa3, 0xe3, 0x1b, 0x33, 0x03, 0x29, 0x4b, 0x12, 0x4c, 0x1e, 0xe6, 0x5e, 0x0e, 0x6c, 0xa1, 0xb9, 0x36, 0xfa, 0x6c, 0x7f, 0xc6, 0xa8, 0x38, 0x73, 0x2a, 0x0a, 0x25, 0x69, 0xa5, 0x97, 0x3f, 0x24, 0x00, 0x30, 0x4d, 0x27, 0xb3, 0x94, 0x48, 0xef, 0x47, 0x98, 0x71, 0x0d, 0x56, 0x76, 0xec, 0x41, 0x12, 0x9b, 0x7b, 0x9c, 0xf5, 0x85, 0x07, 0x2d, 0x6b, 0xc6, 0xc1, 0x2e, 0x72, 0x22, 0x5a, 0x43, 0xff, 0x1e, 0xec, 0x67, 0x2b, 0x31, 0x96, 0x14, 0x2c, 0xb1, 0x5f, 0x5d, 0x0c, 0xc9, 0xad, 0x15, 0x5f, 0xab, 0x66, 0x14, 0x1c, 0x72, 0xfa, 0x23, 0xef, 0x9f, 0x77, 0xf6, 0x50, 0xb0, 0x70, 0xb8, 0x3c, 0x85, 0x9e, 0x90, 0x69, 0x17, 0x25, 0xae, 0x6e, 0xe2, 0x16, 0x7d, 0x42, 0x38, 0xdf, 0x74, 0x72, 0x7b, 0x97, 0xa9, 0x9e, 0x40, 0x24, 0x85, 0xdc, 0x64, 0xfa, 0xb1, 0x8b, 0x95, 0xe6, 0xe4, 0x13, 0x72, 0xf1, 0x52, 0x2f, 0xa0, 0xd6, 0x52, 0xc0, 0x11, 0xa7, 0xfe, 0xd5, 0x3b, 0x56, 0xca, 0xbc, 0x01, 0xce, 0x3d, 0xd2, 0x30, 0x97, 0x1d, 0xdc, 0xeb, 0x9d, 0xa9, 0x3e, 0x09, 0xef, 0xee, 0x7f, 0x09, 0x7b, 0x82, 0x43, 0x15, 0x2e, 0xa4, 0x2e, 0x97, 0x21, 0x92, 0x7e, 0x69, 0x21, 0x25, 0xda, 0x46, 0x7c, 0x0c, 0xcd, 0x1d, 0xde, 0x42, 0x11, 0xa2, 0xef, 0xa2, 0xc8, 0x32, 0x9a, 0x82, 0xcf, 0x72, 0x7e, 0x22, 0xa6, 0x11, 0xfa, 0xec, 0x0b, 0x77, 0x99, 0x38, 0x03, 0xf6, 0x80, 0xba, 0xea, 0x75, 0x19, 0xb0, 0x48, 0x02, 0xb2, 0x6b, 0xc0, 0x8c, 0xfb, 0xfe, 0xaf, 0x94, 0x4f, 0x6f, 0xb4, 0xcb, 0x1c, 0x27, 0xf0, 0x41, 0xb6, 0x46, 0x41, 0x68, 0x3d, 0x05, 0x79, 0x6b, 0xcd, 0xb7, 0x20, 0xdc, 0x40, 0x81, 0x58, 0xcb, 0x33, 0xa3, 0xf3, 0x34, 0xdc, 0x63, 0x2d, 0xa5, 0xb5, 0xa1, 0xd1, 0xfd, 0x49, 0x5b, 0x46, 0x94, 0x01, 0xa8, 0xf2, 0xd8, 0x93, 0x2c, 0xbb, 0x57, 0xfe, 0x7c, 0x77, 0x3b, 0x19, 0x6f, 0x3c, 0xaa, 0x23, 0x5b, 0xc0, 0xe7, 0x00, 0x41, 0x97, 0x91, 0xe8, 0x00, 0x12, 0xdf, 0xf6, 0x5c, 0x2e, 0xc6, 0x8e, 0xc6, 0x77, 0x59, 0x78, 0x9b, 0xef, 0x63, 0xb0, 0xd7, 0xbb, 0xc4, 0x0b, 0x60, 0x65, 0x3f, 0xfe, 0xbf, 0x04, 0x3e, 0xae, 0xc2, 0xa5, 0x90, 0xe1, 0x2a, 0x56, 0x3f, 0x4c, 0x3f, 0x7a, 0x7d, 0xda, 0x81, 0x50, 0xea, 0x4c, 0xfe, 0xc3, 0xf8, 0x5c, 0x2b, 0x67, 0xb3, 0x9f, 0x8b, 0x95, 0xda, 0x6f, 0x5d, 0xdd, 0x82, 0x7f, 0x52, 0xa2, 0xcc, 0x57, 0xec, 0xc4, 0x14, 0xd2, 0x4f, 0x1b, 0xcb, 0xea, 0xaf, 0x0e, 0x0f, 0x53, 0xaa, 0x56, 0x63, 0xea, 0x36, 0xa6, 0x89, 0x1a, 0x66, 0xc0, 0x4e, 0xf4, 0x1e, 0x02, 0x43, 0xde, 0xde, 0xc8, 0x9e, 0x88, 0x6e, 0x32, 0xd4, 0xcb, 0x47, 0x24, 0x7c, 0x28, 0x38, 0xd4, 0x95, 0xb6, 0xa3, 0x91, 0x69, 0xc7, 0x8d, 0xfd, 0x15, 0xf5, 0xbf, 0xb1, 0x98, 0x8c, 0x57, 0x51, 0xbf, 0x83, 0x6a, 0x35, 0x10, 0x03, 0x50, 0xe5, 0xf7, 0xfa, 0xf8, 0xa5, 0xb0, 0xdb, 0xfb, 0x42, 0x93, 0xbb, 0x17, 0xf7, 0x36, 0xbe, 0x26, 0x66, 0x61, 0xe2
+
+db 0xaf, 0xe0, 0x22, 0x25, 0x2a, 0xae, 0x78, 0x67, 0x8f, 0x7e, 0x9e, 0x59, 0xd7, 0xa3, 0x71, 0xcc, 0x43, 0x85, 0x09, 0xf9, 0x18, 0x52, 0x7b, 0x01, 0x73, 0xcb, 0x31, 0x18, 0x66, 0x79, 0x67, 0x10, 0x67, 0xd8, 0xdf, 0x43, 0xaf, 0x2d, 0x9a, 0x09, 0x9c, 0xd1, 0x37, 0x7e, 0xf5, 0x1c, 0x3c, 0x4f, 0x15, 0xe1, 0x6f, 0xfd, 0x13, 0x3d, 0x53, 0x81, 0xa9, 0x93, 0x5f, 0x92, 0x41, 0x48, 0xec, 0x87, 0x87, 0x1d, 0x0b, 0xaa, 0xaa, 0xd3, 0xc2, 0x98, 0x20, 0xce, 0x28, 0xaf, 0x9d, 0x84, 0x69, 0x4a, 0xfd, 0xc0, 0x9c, 0x2e, 0x50, 0x20, 0xb2, 0x00, 0xc1, 0x81, 0x2a, 0x32, 0x8e, 0x95, 0x20, 0xa7, 0xca, 0x39, 0x28, 0x12, 0x23, 0x0e, 0x43, 0xd3, 0x82, 0x76, 0x73, 0x3c, 0xbf, 0xa9, 0x98, 0xf6, 0x39, 0x6d, 0xd9, 0x15, 0x33, 0x1e, 0x07, 0x7c, 0x08, 0x12, 0x23, 0xbd, 0xd3, 0x34, 0x2d, 0x9a, 0x23, 0x21, 0x46, 0xf3, 0x9a, 0x04, 0x25, 0x62, 0xeb, 0x7e, 0x9a, 0xaa, 0xb6, 0x26, 0xaa, 0x85, 0x01, 0x3a, 0xd8, 0xfc, 0x57, 0x98, 0xb9, 0xe4, 0xc4, 0xe9, 0x11, 0x3e, 0x22, 0x95, 0x3b, 0x41, 0x2b, 0x02, 0x04, 0x6c, 0x75, 0xa5, 0xf2, 0xaa, 0x09, 0x9e, 0x6f, 0xab, 0x1d, 0x2a, 0x5c, 0xde, 0x21, 0xb1, 0x96, 0x2d, 0x86, 0x3f, 0xd0, 0x07, 0x18, 0x1f, 0x87, 0xc2, 0x8f, 0xdf, 0x6a, 0x57, 0x6d, 0x3f, 0x80, 0xc5, 0x08, 0x19, 0xa5, 0x09, 0x65, 0x3d, 0xdc, 0x9e, 0x80, 0x3c, 0x2a, 0x0e, 0x7a, 0x40, 0x04, 0x0b, 0xcc, 0x61, 0xdb, 0x73, 0xfc, 0xa5, 0x0a, 0x42, 0x18, 0xc1, 0xd5, 0xbd, 0x18, 0x78, 0xa1, 0xe4, 0xde, 0x44, 0xec, 0x79, 0xb0, 0x27, 0xaa, 0x45, 0x21, 0x57, 0x19, 0x75, 0x09, 0x5c, 0x58, 0xd5, 0xb9, 0x6f, 0x3b, 0x48, 0x59, 0x41, 0x3e, 0xfd, 0x17, 0x43, 0x27, 0xc3, 0x8d, 0x76, 0x8e, 0x38, 0x47, 0xe7, 0xd2, 0xea, 0x54, 0x73, 0x8a, 0x65, 0x4c, 0x49, 0x91, 0xaf, 0x29, 0x65, 0x0d, 0x81, 0xa4, 0x77, 0xd7, 0x32, 0xd0, 0x69, 0xd9, 0x6b, 0xa3, 0x9b, 0x24, 0xd6, 0x0a, 0xd2, 0x77, 0x38, 0x59, 0x0b, 0xc8, 0x5c, 0xc7, 0x0b, 0x1d, 0xd1, 0xfa, 0xa7, 0x45, 0x3c, 0xeb, 0x5c, 0x8e, 0x25, 0x35, 0x81, 0x6d, 0x6d, 0xfe, 0xb4, 0x63, 0x89, 0xe4, 0xf0, 0xa8, 0xda, 0xb7, 0xd4, 0xff, 0x5d, 0x28, 0x97, 0x11, 0xf9, 0x8d, 0xab, 0x29, 0xd5, 0xd3, 0x1c, 0x70, 0x20, 0x4c, 0x41, 0x16, 0x42, 0xfd, 0xfc, 0x62, 0x82, 0x40, 0x59, 0x34, 0x28, 0xd0, 0xd5, 0xfc, 0xac, 0x97, 0xb8, 0x82, 0x0e, 0x4b, 0xae, 0x51, 0x28, 0x1a, 0xf1, 0x87, 0xd3, 0x20, 0xa3, 0xe7, 0x74, 0x69, 0x3c, 0x54, 0x8d, 0xc5, 0x56, 0x1d, 0xcd, 0x75, 0xae, 0x88, 0x17, 0x30, 0xdf, 0x46, 0x4a, 0xbc, 0x64, 0xff, 0xa2, 0xc1, 0x62, 0xbd, 0x88, 0x7b, 0x3e, 0xa1, 0x0c, 0xa9, 0x13, 0x0e, 0xc1, 0xb4, 0x24, 0xe6, 0x96, 0x1b, 0x9c, 0x9b, 0xac, 0x44, 0x33, 0x5b, 0xda, 0xd5, 0x88, 0x4d, 0xfe, 0x81, 0x09, 0x07, 0x17, 0xcf, 0x14, 0x05, 0xaf, 0xf8, 0x72, 0x14, 0x49, 0x5f, 0x06, 0x62, 0xab, 0xe0, 0x42, 0x70, 0x12, 0x59, 0x41, 0x0f, 0x18, 0x83, 0x68, 0x6d, 0xc6, 0x3c, 0xea, 0xe0, 0x6d, 0xd4, 0xae, 0xa6, 0xf1, 0x63, 0x21, 0x7f, 0xb5, 0x9d, 0x22, 0xf4, 0xd2, 0x49, 0x49, 0xed, 0x07, 0xb1, 0x11, 0xf9, 0x2e, 0x74, 0xbe, 0x35, 0x47, 0xdc, 0xef, 0x85, 0x0b, 0x4d, 0x46, 0xe6, 0x1f, 0x60, 0x6a, 0xa1, 0x8a, 0x4d, 0x46, 0x87, 0x30, 0x8e, 0x9a, 0xba, 0x97, 0x3e, 0x15, 0xb7, 0x33, 0x76, 0x81, 0x69, 0xdb, 0x82, 0x5e, 0xe6, 0x7b, 0xec, 0xd2, 0x80, 0x7f, 0x17, 0xf1, 0x73, 0xe2
+
+.data_mid:
+db 0xd7, 0xa9, 0x35, 0x61, 0x5f, 0xb8, 0xb6, 0x2d, 0x29, 0x34, 0x63, 0xbf, 0xe2, 0x1c, 0x34, 0xe9, 0xf5, 0xff, 0x34, 0x8a, 0x2f, 0xea, 0xd4, 0x3f, 0x3b, 0xfe, 0x6e, 0xdf, 0xa6, 0xd8, 0xc6, 0xb8, 0xc5, 0xff, 0x12, 0x97, 0x53, 0xda, 0x86, 0xa8, 0x0a, 0x12, 0x4e, 0x5d, 0x96, 0x65, 0x51, 0x22, 0xe2, 0x9d, 0x08, 0x71, 0x84, 0x19, 0x8b, 0xbf, 0x29, 0xd3, 0x3f, 0xab, 0xde, 0xe4, 0x27, 0x8b, 0x99, 0xcc, 0xb1, 0x7c, 0xa5, 0x71, 0x91, 0x9a, 0x0b, 0xad, 0x75, 0x86, 0xe3, 0x9c, 0x4e, 0x0c, 0x01, 0xb3, 0x12, 0x33, 0x90, 0x81, 0x7c, 0x71, 0x2c, 0x70, 0x61, 0xd5, 0x39, 0x0c, 0x45, 0xfc, 0x27, 0xaf, 0xbb, 0xd9, 0x26, 0x1b, 0x33, 0xb4, 0x0d, 0xf8, 0xd6, 0x2d, 0x09, 0xc7, 0x8c, 0xbf, 0x48, 0x53, 0x14, 0x94, 0x76, 0x25, 0xc7, 0x0c, 0x69, 0x49, 0x82, 0xb4, 0x2f, 0x48, 0x38, 0x44, 0x9d, 0x90, 0x6d, 0x66, 0x35, 0xe9, 0x3e, 0x2f, 0x2a, 0xb7, 0xe1, 0xb1, 0x2b, 0x99, 0x08, 0x6f, 0x5c, 0x6c, 0xdf, 0xdb, 0x10, 0xe2, 0xaa, 0x86, 0xe7, 0xf8, 0x9e, 0x62, 0xde, 0xa5, 0x81, 0x6b, 0x20, 0x47, 0xa9, 0x06, 0x49, 0xc0, 0x78, 0x8c, 0x70, 0x93, 0x7e, 0xda, 0xda, 0x5e, 0x3b, 0x23, 0xf9, 0xcc, 0x87, 0xdf, 0x48, 0x4f, 0xd6, 0x77, 0xce, 0x45, 0xe1, 0xdc, 0x0c, 0x7a, 0x0c, 0x50, 0x15, 0x63, 0x8c, 0x48, 0xd3, 0x8e, 0xfa, 0xcc, 0xac, 0x1a, 0x83, 0xde, 0xb1, 0x87, 0x2a, 0x58, 0x5c, 0xa5, 0x20, 0x3d, 0xaa, 0x1e, 0x5d, 0x71, 0xa6, 0x57, 0x75, 0x82, 0xb7, 0x33, 0x9e, 0x6b, 0xf3, 0x35, 0x02, 0x98, 0x03, 0xe1, 0x3b, 0xd2, 0x9f, 0x7a, 0x06, 0x85, 0xef, 0x7d, 0xd9, 0xf2, 0x0c, 0x9e, 0xce, 0xb9, 0xce, 0x13, 0x4a, 0x9e, 0x8a, 0x29, 0xe6, 0xe5, 0xe4, 0x39, 0xba, 0xfd, 0xa3, 0x33, 0xa8, 0x13, 0x9e, 0xa5, 0x11, 0x37, 0x69, 0xbc, 0xda, 0x11, 0x49, 0x2d, 0x4a, 0xef, 0x20, 0x8b, 0x7a, 0xb8, 0x9c, 0xc3, 0xaf, 0x26, 0x71, 0xd9, 0xa2, 0xf6, 0x0f, 0x85, 0x87, 0xa8, 0x6c, 0xf9, 0x99, 0xa2, 0xb2, 0x36, 0x2d, 0x78, 0x10, 0xe4, 0x33, 0x8d, 0xa4, 0x63, 0xea, 0x02, 0xb9, 0xac, 0x2f, 0x90, 0x39, 0x2d, 0x0e, 0x2e, 0xf5, 0x08, 0xa5, 0x5c, 0x8e, 0x71, 0x30, 0x0d, 0x1b, 0x84, 0x7a, 0xd7, 0xd4, 0xab, 0x81, 0x82, 0x18, 0x37, 0xf3, 0x28, 0x6f, 0x4e, 0x28, 0x71, 0xda, 0xc9, 0x99, 0x46, 0x14, 0x46, 0x77, 0x01, 0x16, 0x21, 0xae, 0x83, 0x93, 0x86, 0x7f, 0x5a, 0xee, 0xd5, 0xdf, 0x48, 0x5b, 0x15, 0xc8, 0x09, 0x30, 0x8f, 0x01, 0xcc, 0x95, 0x30, 0xd9, 0xf7, 0x72, 0x97, 0xfd, 0x9d, 0xec, 0x9f, 0xbf, 0x5c, 0xbf, 0x4f, 0xca, 0x33, 0xb4, 0xd2, 0xa2, 0xb9, 0x08, 0x9c, 0x40, 0x25, 0x3f, 0x86, 0xdc, 0x83, 0x70, 0x2f, 0xfb, 0x2a, 0xf8, 0x61, 0x1f, 0xa1, 0x1f, 0x36, 0x04, 0xe2, 0xef, 0x1c, 0xa4, 0xcd, 0x3c, 0x7f, 0xc5, 0x73, 0x9c, 0x2e, 0xeb, 0x03, 0x79, 0xd1, 0x02, 0xfc, 0x6f, 0xbd, 0x5a, 0x95, 0xb2, 0xf6, 0x25, 0x96, 0xe6, 0x80, 0x0a, 0xc5, 0xc7, 0xca, 0x8d, 0x31, 0xae, 0xf0, 0x49, 0xcf, 0x43, 0x06, 0x27, 0x7f, 0x25, 0xc7, 0x4c, 0xb7, 0xfc, 0x73, 0xd3, 0x04, 0xd3, 0xb9, 0x9f, 0x74, 0xed, 0x9e, 0x3c, 0xf0, 0xcf, 0x26, 0x2b, 0xd9, 0xcb, 0x78, 0x2a, 0xef, 0x72, 0xf7, 0xb6, 0x78, 0x30, 0x2d, 0x8c, 0x83, 0x73, 0x66, 0x74, 0x3d, 0x66, 0x0a, 0x74, 0x5a, 0x3f, 0x9f, 0x6e, 0x56, 0x68, 0x01, 0xc2, 0xca, 0x2b, 0xa1, 0x25, 0x36, 0x9c, 0x3b, 0xa4, 0x5e, 0x44, 0xf1, 0x18, 0x1d, 0xb6, 0x1a, 0x3a, 0xee, 0x8d, 0x67, 0x34, 0x9c
+
+db 0xdd, 0x48, 0x14, 0xc2, 0x5f, 0xd8, 0xe5, 0x71, 0x22, 0xbf, 0xbc, 0x84, 0xda, 0xc1, 0xb1, 0x22, 0x55, 0xa4, 0x63, 0x41, 0x77, 0xac, 0x40, 0x2d, 0x44, 0x73, 0x8c, 0x14, 0xba, 0x5e, 0x63, 0x68, 0x65, 0x61, 0x6d, 0xec, 0xe2, 0x6d, 0x37, 0x22, 0x04, 0xeb, 0xc7, 0xd4, 0xc9, 0x62, 0x56, 0x13, 0x96, 0x29, 0x03, 0xf4, 0x55, 0xe2, 0x58, 0x7d, 0xda, 0x52, 0x2e, 0x94, 0x07, 0xe6, 0xef, 0xc0, 0xee, 0x9e, 0x0b, 0xf7, 0xcd, 0x13, 0x8b, 0x7d, 0xea, 0xdc, 0xf8, 0xf1, 0xcb, 0xad, 0x49, 0x97, 0xc9, 0x98, 0x0b, 0xcf, 0x84, 0x8e, 0x8e, 0xbb, 0x06, 0x2e, 0x54, 0xf5, 0xa7, 0xbd, 0x70, 0x7e, 0x38, 0x69, 0x8d, 0xb0, 0x01, 0x7b, 0x41, 0x80, 0x09, 0x44, 0xfd, 0x7e, 0x21, 0xb4, 0xbe, 0x6b, 0x4a, 0xb7, 0xca, 0x2d, 0x19, 0xfe, 0x6d, 0xd6, 0x11, 0x29, 0xbb, 0xb2, 0x16, 0xf1, 0xe7, 0x92, 0x71, 0xda, 0x7e, 0x68, 0x3a, 0xe0, 0xea, 0x89, 0x8d, 0xe0, 0x44, 0x48, 0x25, 0x92, 0x37, 0x54, 0x26, 0xf2, 0xab, 0xb3, 0x3b, 0xdb, 0xbb, 0x2b, 0x5c, 0xf5, 0xbc, 0xc7, 0x97, 0xdb, 0xc7, 0x49, 0x25, 0x7c, 0xc2, 0x80, 0x02, 0x69, 0xd4, 0xda, 0xda, 0xe1, 0x04, 0xf3, 0x19, 0xb8, 0xc9, 0xb2, 0xfb, 0x1e, 0x47, 0xa9, 0x0c, 0xa3, 0x48, 0xce, 0xc2, 0x9e, 0x3b, 0x28, 0x23, 0x5a, 0x20, 0x44, 0x77, 0x40, 0xe2, 0xd7, 0x20, 0xd5, 0x71, 0x6f, 0xd4, 0x3c, 0x68, 0x38, 0x9b, 0x89, 0x2e, 0x2d, 0xa8, 0x1f, 0x99, 0xb5, 0x8a, 0x66, 0x07, 0x59, 0x75, 0x9e, 0xf8, 0xd9, 0xbe, 0x85, 0x6a, 0x20, 0x92, 0x9d, 0xd2, 0x5e, 0x45, 0xc0, 0x60, 0xbe, 0x85, 0x0b, 0x84, 0x47, 0xf5, 0xa8, 0x43, 0x87, 0xf1, 0x21, 0x21, 0xb0, 0x3b, 0x04, 0x13, 0x16, 0x3e, 0xdf, 0xc3, 0xc6, 0x04, 0x73, 0xcd, 0x92, 0x76, 0xfb, 0xe7, 0x9c, 0xd3, 0x46, 0x11, 0x78, 0xca, 0x12, 0xd9, 0x4a, 0x35, 0xf1, 0x6e, 0x89, 0x8b, 0xe9, 0x7a, 0x04, 0xba, 0x18, 0x25, 0x7c, 0x9e, 0xe6, 0x4f, 0xc2, 0x56, 0x05, 0x72, 0xc3, 0x76, 0xee, 0x7d, 0x77, 0x19, 0x7a, 0x73, 0x2c, 0x81, 0xb8, 0xc7, 0xd9, 0x7f, 0x17, 0x5d, 0x30, 0xda, 0x77, 0x3c, 0x14, 0x88, 0xe8, 0xe4, 0xbf, 0xee, 0x21, 0x1c, 0x29, 0x4e, 0x58, 0xa8, 0x8a, 0x5c, 0xae, 0xa2, 0x1c, 0x7c, 0x25, 0x7c, 0x1c, 0x39, 0xa4, 0x28, 0x4b, 0x78, 0x52, 0xae, 0x2c, 0xbb, 0x5f, 0xbf, 0x51, 0x09, 0x20, 0x76, 0xb2, 0x7d, 0xb1, 0x63, 0x84, 0xc5, 0x49, 0x8a, 0x73, 0xdb, 0x76, 0x1d, 0x25, 0x31, 0xf2, 0x1e, 0x19, 0x38, 0xc8, 0x3b, 0x51, 0x3c, 0x13, 0x52, 0x84, 0xae, 0xc2, 0xe4, 0x8a, 0x57, 0x0d, 0xde, 0x8d, 0x18, 0x48, 0x9a, 0xbd, 0xbf, 0xf3, 0xea, 0x79, 0x17, 0x06, 0x96, 0x72, 0x08, 0x60, 0x95, 0xf9, 0x6f, 0x25, 0x0c, 0xb7, 0x9d, 0x98, 0x23, 0x01, 0xc8, 0x7a, 0xdb, 0x75, 0x63, 0x64, 0x14, 0x5e, 0x10, 0xf5, 0x16, 0x48, 0xbc, 0xc6, 0x7e, 0x24, 0xf3, 0xad, 0x57, 0x3f, 0x7d, 0x6c, 0xab, 0x18, 0x8c, 0x12, 0xc5, 0x0c, 0xd8, 0xb5, 0x1e, 0x43, 0x7c, 0x23, 0x17, 0x48, 0xba, 0x76, 0x3b, 0xd9, 0x2b, 0xae, 0x1b, 0xef, 0x58, 0xfa, 0x87, 0xad, 0x9b, 0x6d, 0xf9, 0xab, 0xa8, 0x3c, 0xfc, 0x59, 0x67, 0xa6, 0x2c, 0xc7, 0x75, 0xa4, 0x97, 0xca, 0x18, 0x18, 0x04, 0x2c, 0xb3, 0x0e, 0xa9, 0x69, 0x33, 0x67, 0xa2, 0xc6, 0xbc, 0x98, 0x48, 0x71, 0x11, 0x05, 0x30, 0xf6, 0xa9, 0x61, 0x40, 0x46, 0xf1, 0x41, 0x37, 0xd0, 0x6b, 0x7c, 0x1f, 0x03, 0x5c, 0xe9, 0xf4, 0x59, 0x1d, 0x35, 0xf0, 0x98, 0x42, 0x4a, 0x92, 0x2a, 0xc3, 0x9a, 0xb8, 0xa5