feat(relinker): --to-intel lowers SSE4a sites to same-length jumps into an appended stub section

EXTRQ, INSERTQ, MOVNTSS and MOVNTSD are rewritten at relink time: same-length in-place forms
where they exist, otherwise a jmp rel32 into an appended executable section holding a
scratch-free SSE body and a jump back, so no code shifts and no RIP-relative or jump-table
operand needs adjusting. Decoder fixes for POPCNT and 0F BB (BTC), codegen and elfpatcher
tests, README and technical-debt notes. The runtime illegal-instruction trap stays as a
fallback.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_0177gthrWEBBfBu5xJDXgBfo

core/relinker/codegen/include/codegen/CodegenTypes.hpp

@@ -1,6 +1,7 @@
 #ifndef CODEGEN_CODEGENTYPES_HPP
 #define CODEGEN_CODEGENTYPES_HPP
 
+#include <codegen/x86/Amd64OnlySubstitutionTypes.hpp>
 #include <domain/Types.hpp>
 #include <cstdint>
 #include <vector>
@@ -28,9 +29,20 @@ struct RewriteResult {
     std::vector<AddressAdjustment> Adjustments;
 };
 
+struct TrampolineSite {
+    Domain::FileByteOffset Offset;
+    Domain::VirtualAddress Address;
+    std::size_t Length;
+    std::vector<std::uint8_t> OriginalBytes;
+    std::vector<std::uint8_t> Body;
+    std::size_t ReturnBranchOffset;
+};
+
 struct ConvertResult {
     std::vector<std::uint8_t> Bytes;
     std::size_t ReplacedCount;
+    std::vector<TrampolineSite> Trampolines;
+    std::vector<Amd64OnlySubstitutionReport> Reports;
 };
 
 enum class ControlFlowKind : std::uint8_t {

core/relinker/codegen/include/codegen/x86/Amd64OnlySubstitutionTable.hpp

@@ -17,14 +17,39 @@ struct Entry {
     inline constexpr std::uint8_t k##name##Bytes[] = {__VA_ARGS__}; \
     inline constexpr Entry k##name = {#name, k##name##Bytes, sizeof(k##name##Bytes)};
 
-AMD64_STUB(Monitorx, 0x0F, 0x01, 0xFA)
-AMD64_STUB(Mwaitx, 0x0F, 0x01, 0xFB)
-AMD64_STUB(Clzero, 0x0F, 0x01, 0xFC)
-AMD64_STUB(Rdpru, 0x0F, 0x01, 0xFD)
-AMD64_STUB(Mcommit, 0xF3, 0x0F, 0x01, 0xFA)
+AMD64_STUB(JmpRel32, 0xE9, 0x00, 0x00, 0x00, 0x00)
+AMD64_STUB(LeaRspBelowRedZone, 0x48, 0x8D, 0xA4, 0x24, 0x70, 0xFF, 0xFF, 0xFF)
+AMD64_STUB(LeaRspRestore, 0x48, 0x8D, 0xA4, 0x24, 0x90, 0x00, 0x00, 0x00)
+AMD64_STUB(Nop1, 0x90)
+AMD64_STUB(Nop2, 0x66, 0x90)
+AMD64_STUB(Nop3, 0x0F, 0x1F, 0x00)
+AMD64_STUB(Nop4, 0x0F, 0x1F, 0x40, 0x00)
+AMD64_STUB(Nop5, 0x0F, 0x1F, 0x44, 0x00, 0x00)
+AMD64_STUB(Nop6, 0x66, 0x0F, 0x1F, 0x44, 0x00, 0x00)
+AMD64_STUB(Nop7, 0x0F, 0x1F, 0x80, 0x00, 0x00, 0x00, 0x00)
 
 #undef AMD64_STUB
 
+inline constexpr Entry kNops[] = {kNop1, kNop2, kNop3, kNop4, kNop5, kNop6, kNop7};
+
+inline constexpr Entry kMonitorx = {"MONITORX", nullptr, 0};
+inline constexpr Entry kMwaitx = {"MWAITX", nullptr, 0};
+inline constexpr Entry kClzero = {"CLZERO", nullptr, 0};
+inline constexpr Entry kRdpru = {"RDPRU", nullptr, 0};
+inline constexpr Entry kMcommit = {"MCOMMIT", nullptr, 0};
+inline constexpr Entry kExtrq = {"EXTRQ", nullptr, 0};
+inline constexpr Entry kInsertq = {"INSERTQ", nullptr, 0};
+inline constexpr Entry kExtrqRegisterForm = {"EXTRQ register form", nullptr, 0};
+inline constexpr Entry kInsertqRegisterForm = {"INSERTQ register form", nullptr, 0};
+inline constexpr Entry kMovntss = {"MOVNTSS", nullptr, 0};
+inline constexpr Entry kMovntsd = {"MOVNTSD", nullptr, 0};
+
+inline constexpr std::uint8_t kMovsStoreOpcode = 0x11;
+inline constexpr std::uint8_t kPshufbZero = 0x80;
+inline constexpr std::uint8_t kTrapFill = 0xCC;
+inline constexpr std::size_t kStubAlignment = 16;
+inline constexpr std::size_t kRedZoneSpillFrame = 0x90;
+
 } // namespace Amd64OnlySubstitutionTable
 } // namespace Codegen
 

core/relinker/codegen/include/codegen/x86/Amd64OnlySubstitutionTypes.hpp

@@ -8,10 +8,19 @@
 
 namespace Codegen {
 
+enum class Amd64OnlyLowering : std::uint8_t {
+    InPlace,
+    Trampoline,
+    Unsupported
+};
+
 struct Amd64OnlyMatch {
     std::string InstructionName;
     std::size_t Length;
+    Amd64OnlyLowering Lowering;
     std::vector<std::uint8_t> ReplacementBytes;
+    std::vector<std::uint8_t> StubBody;
+    std::size_t ReturnBranchOffset;
 };
 
 struct Amd64OnlySubstitutionReport {
@@ -19,6 +28,7 @@ struct Amd64OnlySubstitutionReport {
     Domain::FileByteOffset Offset;
     std::size_t OriginalLength;
     std::size_t ReplacementLength;
+    Amd64OnlyLowering Lowering;
 };
 
 }

core/relinker/codegen/include/codegen/x86/DecodedInstruction.hpp

@@ -10,6 +10,7 @@ struct DecodedInstruction {
     const std::uint8_t* Data;
     std::size_t Length;
 
+    [[nodiscard]] std::size_t OpcodeOffset() const;
     [[nodiscard]] bool IsShaNi() const;
     [[nodiscard]] bool IsExtrq() const;
     [[nodiscard]] bool IsInsertq() const;

core/relinker/codegen/include/codegen/x86/Sse4aLowering.hpp

@@ -0,0 +1,25 @@
+#ifndef CODEGEN_X86_SSE4ALOWERING_HPP
+#define CODEGEN_X86_SSE4ALOWERING_HPP
+
+#include <codegen/x86/Sse4aOperands.hpp>
+#include <cstddef>
+#include <cstdint>
+#include <optional>
+#include <vector>
+
+namespace Codegen {
+
+struct LoweredBody {
+    std::vector<std::uint8_t> Bytes;
+    std::size_t ReturnBranchOffset;
+};
+
+class Sse4aLowering {
+public:
+    [[nodiscard]] std::optional<std::vector<std::uint8_t>> LowerInPlace(const Sse4aOperands& operands, std::size_t originalLength) const;
+    [[nodiscard]] LoweredBody LowerOutOfLine(const Sse4aOperands& operands) const;
+};
+
+}
+
+#endif

core/relinker/codegen/include/codegen/x86/Sse4aOperands.hpp

@@ -0,0 +1,22 @@
+#ifndef CODEGEN_X86_SSE4AOPERANDS_HPP
+#define CODEGEN_X86_SSE4AOPERANDS_HPP
+
+#include <cstddef>
+#include <cstdint>
+
+namespace Codegen {
+
+struct Sse4aOperands {
+    bool Insertq;
+    bool RegisterForm;
+    std::uint8_t Destination;
+    std::uint8_t Source;
+    std::uint8_t Length;
+    std::uint8_t Index;
+};
+
+[[nodiscard]] Sse4aOperands DecodeSse4a(const std::uint8_t* data, std::size_t length);
+
+}
+
+#endif

core/relinker/codegen/include/codegen/x86/X64OpcodeConstants.hpp

@@ -106,6 +106,10 @@ inline constexpr std::uint8_t OneByteLoop = 0xE0;
 inline constexpr std::uint8_t OneByteLoopMax = 0xE2;
 inline constexpr std::uint8_t OneByteInOutImm8Min = 0xE4;
 inline constexpr std::uint8_t OneByteInOutImm8Max = 0xE7;
+inline constexpr std::uint8_t OneByteInt = 0xCD;
+inline constexpr std::uint8_t OneByteRetImm16 = 0xC2;
+inline constexpr std::uint8_t OneByteRetFarImm16 = 0xCA;
+inline constexpr std::uint8_t OneByteEnter = 0xC8;
 
 inline constexpr std::uint8_t TwoByteJccRel32Min = 0x80;
 inline constexpr std::uint8_t TwoByteJccRel32Max = 0x8F;
@@ -158,6 +162,19 @@ inline constexpr std::uint8_t TwoByteShufpdImm8 = 0xC6;
 inline constexpr std::uint8_t TwoByteGrp9 = 0xC7;
 inline constexpr std::uint8_t TwoByteExtrqInsertqImm8Imm8 = 0x78;
 inline constexpr std::uint8_t TwoByteExtrqInsertqModRm = 0x79;
+inline constexpr std::uint8_t TwoByteMovnts = 0x2B;
+inline constexpr std::uint8_t TwoByteGrp6 = 0x00;
+inline constexpr std::uint8_t TwoByteLar = 0x02;
+inline constexpr std::uint8_t TwoByteLsl = 0x03;
+inline constexpr std::uint8_t TwoBytePrefetchw = 0x0D;
+inline constexpr std::uint8_t TwoByteMovCrDrMin = 0x20;
+inline constexpr std::uint8_t TwoByteMovCrDrMax = 0x23;
+inline constexpr std::uint8_t TwoBytePopcnt = 0xB8;
+inline constexpr std::uint8_t TwoByteUd1 = 0xB9;
+inline constexpr std::uint8_t TwoByteBtc = 0xBB;
+inline constexpr std::uint8_t TwoByteXaddRm8 = 0xC0;
+inline constexpr std::uint8_t TwoByteMovnti = 0xC3;
+inline constexpr std::uint8_t TwoBytePinsrw = 0xC4;
 
 inline constexpr std::uint8_t VexNoModRmMin = 0x77;
 inline constexpr std::uint8_t VexNoModRmMax = 0x77;
@@ -171,6 +188,7 @@ inline constexpr std::uint8_t EvexMap0F38 = 0x02;
 inline constexpr std::size_t ImmSizeNone = 0;
 inline constexpr std::size_t ImmSize8 = 1;
 inline constexpr std::size_t ImmSize16 = 2;
+inline constexpr std::size_t ImmSizeEnter = 3;
 inline constexpr std::size_t ImmSize32 = 4;
 inline constexpr std::size_t ImmSize64 = 8;