fix(lint): calibrate register-clobber to the Go ABI and add legacy SSE moves

Assisted-by: Qwen 3.8 Max Preview

asm/instrs.go

@@ -661,6 +661,66 @@ func (e *enc) encodeMovExtend(base string, ops []Operand) error {
 	return e.emit(i)
 }
 
+// --- legacy SSE moves --------------------------------------------------------
+
+// sseMove describes a legacy (non-VEX) SSE move: a mandatory prefix plus a
+// load opcode (reg = destination, rm = source) and a store opcode (the
+// reverse).  The Plan 9 names MOVOU/MOVO are the integer unaligned/aligned
+// octa moves (MOVDQU/MOVDQA), not the packed-single ones.
+type sseMove struct {
+	prefix byte // 0, 0x66, 0xF2 or 0xF3
+	load   byte
+	store  byte
+}
+
+var sseMoveTable = map[string]sseMove{
+	"MOVOU":  {0xF3, 0x6F, 0x7F}, // MOVDQU — unaligned octa
+	"MOVO":   {0x66, 0x6F, 0x7F}, // MOVDQA — aligned octa
+	"MOVUPS": {0x00, 0x10, 0x11}, // unaligned packed single
+	"MOVAPS": {0x00, 0x28, 0x29}, // aligned packed single
+	"MOVUPD": {0x66, 0x10, 0x11}, // unaligned packed double
+	"MOVAPD": {0x66, 0x28, 0x29}, // aligned packed double
+	"MOVSD":  {0xF2, 0x10, 0x11}, // scalar double
+	"MOVSS":  {0xF3, 0x10, 0x11}, // scalar single
+}
+
+// encodeSSEMove encodes a legacy SSE move: a vector-to-vector move uses the
+// load form (reg = destination), matching the Go assembler.
+func (e *enc) encodeSSEMove(m sseMove, ops []Operand) error {
+	if len(ops) != 2 {
+		return fmt.Errorf("SSE move expects 2 operands, got %d", len(ops))
+	}
+	src, dst := ops[0], ops[1]
+	srcReg, srcVec := vecReg(src)
+	dstReg, dstVec := vecReg(dst)
+	op := m.store
+	var reg Reg
+	var rm Operand
+	switch {
+	case srcVec && dstVec:
+		op = m.load
+		reg, rm = dstReg, src
+	case srcVec:
+		if _, ok := dst.(Mem); !ok {
+			return fmt.Errorf("SSE move: invalid destination operand")
+		}
+		reg, rm = srcReg, dst
+	case dstVec:
+		if _, ok := src.(Mem); !ok {
+			return fmt.Errorf("SSE move: invalid source operand")
+		}
+		op = m.load
+		reg, rm = dstReg, src
+	default:
+		return fmt.Errorf("SSE move needs a vector register operand")
+	}
+	i := &instr{prefix: m.prefix, opcode: []byte{0x0F, op}, modrm: -1, sib: -1}
+	if err := setRM(i, reg, rm, 8); err != nil {
+		return err
+	}
+	return e.emit(i)
+}
+
 // --- CVTSL2SD / CVTSQ2SD -----------------------------------------------------
 
 // encodeCvtsi2sd encodes a signed integer to scalar double conversion