feat(relinker): lower SHA-256 instructions with a memory operand

SHA256RNDS2/MSG1/MSG2 with a memory source made `--to-intel` fail the
whole relink. The stub now spills a scratch xmm, loads the operand into
it with MOVDQU (AMD does not fault on a misaligned SHA operand, so the
unaligned load keeps that behaviour) and runs the register-form
lowering. RSP-based operands get the spill depth added to their
displacement. RIP-relative operands still fail: the stub runs at a
different address.

Co-Authored-By: Claude <[email protected]>

core/relinker/codegen/include/codegen/x86/Sha256Operands.hpp

@@ -1,8 +1,10 @@
 #ifndef CODEGEN_X86_SHA256OPERANDS_HPP
 #define CODEGEN_X86_SHA256OPERANDS_HPP
 
+#include <codegen/x86/StubBodyBuilder.hpp>
 #include <cstddef>
 #include <cstdint>
+#include <optional>
 
 namespace Codegen {
 
@@ -16,6 +18,7 @@ struct Sha256Operands {
     Sha256Operation Operation;
     std::uint8_t Destination;
     std::uint8_t Source;
+    std::optional<MemoryOperand> Memory;
 };
 
 [[nodiscard]] Sha256Operands DecodeSha256(const std::uint8_t* data, std::size_t length);

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

@@ -17,6 +17,18 @@ struct LoweredBody {
 
 using StubConstant = std::array<std::uint8_t, 16>;
 
+struct MemoryOperand {
+    std::vector<std::uint8_t> Prefixes;
+    std::uint8_t RexIndexBase;
+    std::uint8_t Mod;
+    std::uint8_t Rm;
+    std::uint8_t Sib;
+    std::int32_t Displacement;
+    bool StackBase;
+};
+
+[[nodiscard]] MemoryOperand DecodeMemoryOperand(const std::uint8_t* data, std::size_t length, std::size_t modRmOffset, std::uint8_t rex, std::vector<std::uint8_t> prefixes);
+
 void EmitSse(std::vector<std::uint8_t>& out, std::uint8_t prefix, std::initializer_list<std::uint8_t> opcode, std::uint8_t dst, std::uint8_t src);
 void EmitShiftImm(std::vector<std::uint8_t>& out, std::uint8_t extension, std::uint8_t reg, std::uint8_t imm);
 
@@ -28,6 +40,7 @@ public:
     void ShiftImm(std::uint8_t extension, std::uint8_t reg, std::uint8_t imm);
     void ShiftDwordImm(std::uint8_t extension, std::uint8_t reg, std::uint8_t imm);
     void RipOperand(std::initializer_list<std::uint8_t> opcode, std::uint8_t reg, const StubConstant& constant);
+    void Load(std::uint8_t reg, const MemoryOperand& operand);
     void Spill(std::uint8_t reg);
     void Restore(std::uint8_t reg);
     void Raw(std::span<const std::uint8_t> bytes);
@@ -43,6 +56,7 @@ private:
     std::vector<std::uint8_t> _bytes;
     std::vector<Fixup> _fixups;
     std::vector<StubConstant> _constants;
+    std::size_t _stackDepth = 0;
 };
 
 }