feat(asm): encode the amd64 and loong64 tails of the corpus testdata

Assisted-by: GLM 5.3

asm/assemble.go

@@ -808,17 +808,43 @@ func isJumpMnemonic(mnem string) bool {
 	if mnem == "JMP" || mnem == "CALL" {
 		return true
 	}
+	if isLoopMnemonic(mnem) {
+		return true
+	}
 	_, ok := condCode(mnem)
 	return ok
 }
 
+// isLoopMnemonic reports the LOOP family, rel8 alone (E0-E2).
+func isLoopMnemonic(mnem string) bool {
+	switch mnem {
+	case "LOOP", "LOOPE", "LOOPNE":
+		return true
+	}
+	return false
+}
+
+// loopOpcode maps the LOOP family to its E0-E2 opcode.
+func loopOpcode(mnem string) byte {
+	switch mnem {
+	case "LOOPE":
+		return 0xE1
+	case "LOOPNE":
+		return 0xE0
+	}
+	return 0xE2
+}
+
 // jumpSize returns the length of a jump instruction in the requested form:
 // short (rel8) where available, otherwise the rel32 form.  CALL is always
-// rel32.
+// rel32; the LOOP family is rel8 alone.
 func jumpSize(mnem string, long bool) int {
 	if mnem == "CALL" {
 		return 5 // opcode + rel32
 	}
+	if isLoopMnemonic(mnem) {
+		return 2 // opcode + rel8, the only form
+	}
 	if !long {
 		return 2 // opcode + rel8
 	}
@@ -945,6 +971,16 @@ func encodeNormal(s *ast.Instr, fi frameInfo, link *linkInfo) ([]byte, []sbPatch
 	if (mnemUpper == "MOVQ" || mnemUpper == "MOVL") && len(s.Operands) == 2 && isBareTLS(s.Operands[0]) {
 		return encodeTLSBaseLoad(s, fi, link)
 	}
+	// The old paired-register shift spelling, SHLL CX, R11:AX (a colon
+	// between the two registers), is the toolchain's SHLD family: SHLDL CL,
+	// AX, R11 with the count register first, the paired source in the reg
+	// field and the pair's head in r/m.
+	if code, ps, err := encodeColonShift(s, mnemUpper, fi, link); code != nil || err != nil {
+		if err != nil {
+			return nil, nil, nil, err
+		}
+		return code, ps, nil, nil
+	}
 	_, size := splitSize(mnemUpper)
 	if size == 0 {
 		size = 8
@@ -1051,6 +1087,66 @@ func encodeBookkeeping(upper string, s *ast.Instr) ([]byte, error) {
 	return nil, nil
 }
 
+// encodeColonShift encodes the paired-register shift spellings, SHLx CX,
+// dst:src: the toolchain reads them as the SHLD family (double-precision
+// shift by CL), reg = the paired source, r/m = the pair's head.  The second
+// operand's raw text carries the colon; ok reports the spelling was found.
+func encodeColonShift(s *ast.Instr, mnemUpper string, fi frameInfo, link *linkInfo) ([]byte, []sbPatch, error) {
+	base, _ := strings.CutPrefix(mnemUpper, "SHL")
+	if base == mnemUpper || len(s.Operands) != 2 {
+		return nil, nil, nil
+	}
+	_, size := splitSize(mnemUpper)
+	raw := strings.ReplaceAll(s.Operands[1].Raw, " ", "")
+	head, tail, ok := strings.Cut(raw, ":")
+	if !ok || head == "" || tail == "" {
+		return nil, nil, nil
+	}
+	headReg, ok1 := ParseReg(head)
+	srcReg, ok2 := ParseReg(tail)
+	if !ok1 || !ok2 {
+		return nil, nil, fmt.Errorf("%s: invalid paired register %q", mnemUpper, s.Operands[1].Raw)
+	}
+	cnt, err := operandFromAST(mnemUpper, s.Operands[0], size, fi, link)
+	if err != nil {
+		return nil, nil, err
+	}
+	cntReg, ok := cnt.(Reg)
+	if !ok || cntReg.idx != 1 {
+		return nil, nil, fmt.Errorf("%s: the paired-register form counts in CL", mnemUpper)
+	}
+	// SHLD r/m, reg, CL: 0F A5 (REX.W for the 64-bit width).
+	e := &enc{}
+	i := &instr{rexW: size == 8, opcode: []byte{0x0F, 0xA5}, modrm: -1, sib: -1}
+	if err := setRM(i, srcReg, headReg, size); err != nil {
+		return nil, nil, err
+	}
+	if err := e.emit(i); err != nil {
+		return nil, nil, err
+	}
+	ps := make([]sbPatch, len(e.patches))
+	for j, p := range e.patches {
+		ps[j] = sbPatch{off: p.off, name: p.name, addend: p.addend}
+	}
+	return e.out, ps, nil
+}
+
+// trailingIndexGroup recovers a trailing "(index*scale)" or "(index)" group
+// from an operand's raw text: the symbol-pseudo parse returns before the
+// index group, so foo(SP)(AX*1) keeps its index only in the spelling.
+func trailingIndexGroup(raw string) (string, int, bool) {
+	compact := strings.ReplaceAll(raw, " ", "")
+	if !strings.HasSuffix(compact, ")") {
+		return "", 0, false
+	}
+	open := strings.LastIndex(compact, "(")
+	if open < 2 || !strings.Contains(compact[:open], ")") {
+		return "", 0, false // one group alone: no trailing index
+	}
+	name, scale, _, ok := cutParenGroup(compact[open:])
+	return name, scale, ok
+}
+
 // encodeJump encodes a JMP/CALL/Jcc with a relative offset resolved from the
 // target label or from a numeric ±N(PC) instruction count, in the short
 // (rel8) or long (rel32) form.  numTarget is the resolved byte offset of a
@@ -1085,9 +1181,15 @@ func encodeJump(s *ast.Instr, mnem string, pc int, offsets map[string]int, long
 		if mnem == "JMP" {
 			return []byte{0xEB, byte(int8(rel))}, nil
 		}
+		if isLoopMnemonic(mnem) {
+			return []byte{loopOpcode(mnem), byte(int8(rel))}, nil
+		}
 		cc, _ := condCode(mnem)
 		return []byte{0x70 + byte(cc), byte(int8(rel))}, nil
 	}
+	if isLoopMnemonic(mnem) {
+		return nil, fmt.Errorf("%s has no long form", mnem)
+	}
 	switch mnem {
 	case "JMP":
 		return append([]byte{0xE9}, le32(rel)...), nil
@@ -1139,6 +1241,72 @@ func labelName(op *ast.Operand) (string, bool) {
 	return "", false
 }
 
+// jumpOperand returns the branch-target operand of a JMP/CALL, rewriting the
+// `*`-prefixed indirect spellings (JMP *(R12), JMP *4(SP)) into their plain
+// memory form.  The star marks an indirect target and changes no bytes; the
+// address parser leaves the operand's address empty because of the leading
+// star, so the fields are rebuilt from the raw text onto a copy of the
+// operand, never on the shared syntax tree.
+func jumpOperand(s *ast.Instr) *ast.Operand {
+	if len(s.Operands) != 1 {
+		return nil
+	}
+	op := s.Operands[0]
+	compact := strings.ReplaceAll(op.Raw, " ", "")
+	inner, ok := strings.CutPrefix(compact, "*")
+	if !ok {
+		return op
+	}
+	var addr ast.Address
+	if i := strings.IndexByte(inner, '('); i > 0 {
+		v, err := strconv.ParseInt(inner[:i], 0, 64)
+		if err != nil {
+			return op
+		}
+		addr.Offset, addr.HasOff = v, true
+		inner = inner[i:]
+	}
+	base, _, rest, ok := cutParenGroup(inner)
+	if !ok {
+		return op
+	}
+	if base != "" {
+		addr.Base = base
+	}
+	if rest != "" {
+		idx, scale, _, ok := cutParenGroup(rest)
+		if ok && idx != "" {
+			addr.Index = idx
+			addr.Scale = scale
+		}
+	}
+	c := *op
+	c.Addr = addr
+	return &c
+}
+
+// cutParenGroup splits a leading "(name)" or "(name*n)" off s, returning the
+// inner text, the scale it names (1 when the group spells no multiplier) and
+// the remainder.
+func cutParenGroup(s string) (name string, scale int, rest string, ok bool) {
+	if !strings.HasPrefix(s, "(") {
+		return "", 0, "", false
+	}
+	i := strings.IndexByte(s, ')')
+	if i < 0 {
+		return "", 0, "", false
+	}
+	inner, rest := s[1:i], s[i+1:]
+	if before, after, ok := strings.Cut(inner, "*"); ok {
+		n, err := strconv.Atoi(after)
+		if err != nil {
+			return "", 0, "", false
+		}
+		return before, n, rest, true
+	}
+	return inner, 1, rest, true
+}
+
 // indirectJumpTarget reports whether the JMP/CALL operand addresses a
 // register or a memory location rather than a label or a static symbol.
 // A bare identifier is a register when the register table knows the name and
@@ -1147,7 +1315,14 @@ func indirectJumpTarget(s *ast.Instr) bool {
 	if len(s.Operands) != 1 || s.Operands[0].Kind != ast.OpAddr {
 		return false
 	}
-	a := s.Operands[0].Addr
+	op := jumpOperand(s)
+	if op == nil {
+		return false
+	}
+	if op != s.Operands[0] {
+		return true // the star marker spells an indirect target
+	}
+	a := op.Addr
 	// ±N(PC) is the numeric relative form, the PC counts instructions from
 	// the branch: relative, not indirect.
 	if a.Base == "PC" || a.Index == "PC" {
@@ -1165,18 +1340,20 @@ func indirectJumpTarget(s *ast.Instr) bool {
 }
 
 // encodeIndirectJump assembles a JMP/CALL through a register or memory
-// operand, which carries no relocation and no label to resolve.
+// operand, which carries no relocation and no label to resolve.  The
+// `*`-prefixed spellings go through jumpOperand first, their star rebuilt
+// into a plain memory operand.
 func encodeIndirectJump(s *ast.Instr, mnem string) ([]byte, error) {
-	ops := make([]Operand, len(s.Operands))
-	for i, op := range s.Operands {
-		o, err := operandFromAST(mnem, op, 8, frameInfo{}, nil)
-		if err != nil {
-			return nil, err
-		}
-		ops[i] = o
+	op := s.Operands[0]
+	if cleaned := jumpOperand(s); cleaned != nil {
+		op = cleaned
+	}
+	o, err := operandFromAST(mnem, op, 8, frameInfo{}, nil)
+	if err != nil {
+		return nil, err
 	}
 	e := &enc{}
-	if err := e.encodeIndirectBranch(mnem, ops); err != nil {
+	if err := e.encodeIndirectBranch(mnem, []Operand{o}); err != nil {
 		return nil, err
 	}
 	return e.out, nil
@@ -1245,31 +1422,51 @@ func operandFromAST(mnemUpper string, op *ast.Operand, size int, fi frameInfo, l
 			off := a.Sym.Offset + fi.fpAdjust
 			return Mem{Base: spReg, Disp: off, HasBase: true, Size: size}, nil
 		}
-		// SP-relative local: x-N(SP) → (spAdjust + offset)(SP).
+		// SP-relative local: x-N(SP) → (spAdjust + offset)(SP), keeping a scaled
+		// index beside the virtual stack pointer (foo(SP)(AX*1)).  The
+		// symbol-pseudo parse returns before the index group, so the index
+		// is recovered from the raw text when the address lacks it.
 		if a.Sym != nil && a.Sym.Pseudo == "SP" && a.Base == "" {
 			off := fi.spAdjust + a.Sym.Offset
-			return Mem{Base: spReg, Disp: off, HasBase: true, Size: size}, nil
+			m := Mem{Base: spReg, Disp: off, HasBase: true, Size: size}
+			if name, scale, ok := trailingIndexGroup(op.Raw); ok {
+				idx, ok := ParseReg(name)
+				if !ok {
+					return nil, fmt.Errorf("unknown index register %q", name)
+				}
+				m.Index = idx
+				m.Scale = scale
+				m.HasIndex = true
+			}
+			return m, nil
 		}
 		// SB (global symbol): a symbol defined in the same file (GLOBL) is
 		// encoded RIP-relative and resolved by the file-level layout;
-		// anything not defined here needs object-file emission.
+		// anything not defined here needs object-file emission.  A static
+		// (file-local) spelling of an undefined symbol defers the same way
+		// the toolchain does: the relocation names it and the linker decides.
 		if a.Sym != nil && a.Sym.Pseudo == "SB" {
 			if link == nil || link.symbols == nil {
 				return nil, fmt.Errorf("symbol %q needs file-level assembly (AssembleFile)", a.Sym.Name)
 			}
-			if !link.symbols[a.Sym.Name] {
-				if a.Sym.Static {
-					return nil, fmt.Errorf("undefined symbol %q", a.Sym.Name)
-				}
-				if !link.allowExternal {
-					return nil, fmt.Errorf("external symbol %q needs object-file emission", a.Sym.Name)
-				}
+			if !link.symbols[a.Sym.Name] && !link.allowExternal {
+				return nil, fmt.Errorf("external symbol %q needs object-file emission", a.Sym.Name)
 			}
 			return sbMem{size: size, name: a.Sym.Name, addend: a.Sym.Offset}, nil
 		}
 
 		// Memory with a real base register: (base), off(base), (base)(index*scale).
 		if a.Base != "" {
+			// The TLS pseudo-base, off(TLS): the segment-prefixed absolute
+			// the thread-local access lowers to, 64 8B 04 25 with its
+			// R_TLS_LE patch site on the disp32.
+			if a.Base == "TLS" {
+				seg := byte(0x64) // FS on linux, freebsd, plan9
+				if link != nil && link.goos == "windows" {
+					seg = 0x65 // GS
+				}
+				return TLSMem{Disp: a.Offset, Size: size, Seg: seg}, nil
+			}
 			// Segment-absolute: 0x30(GS) and 0x28(FS), the windows TLS
 			// spellings.  The segment override prefixes a disp32 absolute
 			// reference with no relocation.