feat(asm): assemble static symbols and the whole go-flac AVX2 kernel byte-identically

Assisted-by: Qwen 3.8 Max Preview
2026-07-09 15:56:03 +02:00
parent a82f575aee
commit 56ecc39539
10 changed files with 553 additions and 54 deletions
@@ -23,6 +23,31 @@ import (
 // operands require relocations and are not yet supported; the SIMD (VEX/AVX2)
 // integer and shuffle/extract/permute/move set is in.
 func Assemble(t *ast.Text) ([]byte, map[string]int, error) {
+	code, _, labels, err := assemble(t, nil)
+	return code, labels, err
+}
+
+// linkInfo carries file-level symbol context into a single-function assembly:
+// the set of static symbols a GLOBL in the same file defines.  A nil link
+// rejects SB operands outright (single-function assembly cannot resolve
+// them).
+type linkInfo struct {
+	symbols map[string]bool
+}
+
+// sbPatch is a function-relative static-symbol relocation: the disp32 field
+// at off must become the symbol's address minus after, where after is the
+// function-relative address just past the instruction.
+type sbPatch struct {
+	off    int
+	after  int
+	name   string
+	addend int64
+}
+
+// assemble encodes a TEXT body, returning the machine code, the static-symbol
+// patch sites (for the file-level layout to resolve) and the label table.
+func assemble(t *ast.Text, link *linkInfo) ([]byte, []sbPatch, map[string]int, error) {
 	fi := computeFrame(t)
 	chain := jumpChain(t)
 	resolve := func(name string) string {
@@ -44,9 +69,9 @@ func Assemble(t *ast.Text) ([]byte, map[string]int, error) {
 			case *ast.Label:
 				offsets[s.Name.Text] = pos
 			case *ast.Instr:
-				sz, err := instrSize(s, fi, long[i])
+				sz, err := instrSize(s, fi, long[i], link)
 				if err != nil {
-					return nil, nil, fmt.Errorf("%s: %w", s.Mnemonic.Text, err)
+					return nil, nil, nil, fmt.Errorf("%s: %w", s.Mnemonic.Text, err)
 				}
 				sizes[i] = sz
 				pcs[i] = pos
@@ -85,23 +110,25 @@ func Assemble(t *ast.Text) ([]byte, map[string]int, error) {

 	// Pass 2: emit.
 	out := append([]byte(nil), fi.prologue...)
+	var patches []sbPatch
 	pos := len(fi.prologue)
 	for i, stmt := range t.Body {
 		s, ok := stmt.(*ast.Instr)
 		if !ok {
 			continue
 		}
-		code, err := encodeInstr(s, pos, offsets, fi, long[i], resolve)
+		code, ps, err := encodeInstr(s, pos, offsets, fi, long[i], resolve, link)
 		if err != nil {
-			return nil, nil, fmt.Errorf("%s: %w", s.Mnemonic.Text, err)
+			return nil, nil, nil, fmt.Errorf("%s: %w", s.Mnemonic.Text, err)
 		}
 		if len(code) != sizes[i] {
-			return nil, nil, fmt.Errorf("%s: size mismatch (%d vs %d)", s.Mnemonic.Text, len(code), sizes[i])
+			return nil, nil, nil, fmt.Errorf("%s: size mismatch (%d vs %d)", s.Mnemonic.Text, len(code), sizes[i])
 		}
+		patches = append(patches, ps...)
 		out = append(out, code...)
 		pos += len(code)
 	}
-	return out, offsets, nil
+	return out, patches, offsets, nil
 }

 // jumpChain precomputes jump-to-jump folding: a label whose first instruction
@@ -214,12 +241,12 @@ func addSP(size int) []byte { // ADDQ $size, SP
 // instrSize returns the encoded length of an instruction (layout pass).
 // encodeInstr already includes the epilogue for a RET in a frame-pointer
 // function; jumps use their short or long form (never an epilogue).
-func instrSize(s *ast.Instr, fi frameInfo, long bool) (int, error) {
+func instrSize(s *ast.Instr, fi frameInfo, long bool, link *linkInfo) (int, error) {
 	mnem := strings.ToUpper(s.Mnemonic.Text)
 	if isJumpMnemonic(mnem) {
 		return jumpSize(mnem, long), nil
 	}
-	code, err := encodeInstr(s, 0, nil, fi, false, nil)
+	code, _, err := encodeInstr(s, 0, nil, fi, false, nil, link)
 	if err != nil {
 		return 0, err
 	}
@@ -254,7 +281,7 @@ func jumpSize(mnem string, long bool) int {
 // (relative to pc, the instruction's own offset).  A RET in a frame-pointer
 // function is prefixed with the epilogue.  resolve, when non-nil, redirects a
 // jump label through the jump-to-jump chain before the offset lookup.
-func encodeInstr(s *ast.Instr, pc int, offsets map[string]int, fi frameInfo, long bool, resolve func(string) string) ([]byte, error) {
+func encodeInstr(s *ast.Instr, pc int, offsets map[string]int, fi frameInfo, long bool, resolve func(string) string, link *linkInfo) ([]byte, []sbPatch, error) {
 	mnem := strings.ToUpper(s.Mnemonic.Text)

 	var prefix []byte
@@ -263,32 +290,48 @@ func encodeInstr(s *ast.Instr, pc int, offsets map[string]int, fi frameInfo, lon
 	}

 	var code []byte
+	var ps []sbPatch
 	var err error
 	if isJumpMnemonic(mnem) {
 		code, err = encodeJump(s, mnem, pc+len(prefix), offsets, long, resolve)
 	} else {
-		code, err = encodeNormal(s, fi)
+		code, ps, err = encodeNormal(s, fi, link)
 	}
 	if err != nil {
-		return nil, err
+		return nil, nil, err
 	}
-	return append(prefix, code...), nil
+	// Anchor the patch fields at function-relative positions: off indexes the
+	// disp32 field, after is the address just past the instruction.
+	body := pc + len(prefix)
+	for i := range ps {
+		ps[i].off += body
+		ps[i].after = body + len(code)
+	}
+	return append(prefix, code...), ps, nil
 }

-func encodeNormal(s *ast.Instr, fi frameInfo) ([]byte, error) {
+func encodeNormal(s *ast.Instr, fi frameInfo, link *linkInfo) ([]byte, []sbPatch, error) {
 	_, size := splitSize(strings.ToUpper(s.Mnemonic.Text))
 	if size == 0 {
 		size = 8
 	}
 	ops := make([]Operand, len(s.Operands))
 	for i, op := range s.Operands {
-		o, err := operandFromAST(op, size, fi)
+		o, err := operandFromAST(op, size, fi, link)
 		if err != nil {
-			return nil, err
+			return nil, nil, err
 		}
 		ops[i] = o
 	}
-	return Encode(s.Mnemonic.Text, ops...)
+	e := &enc{}
+	if err := e.encode(s.Mnemonic.Text, ops); err != nil {
+		return nil, nil, err
+	}
+	ps := make([]sbPatch, len(e.patches))
+	for i, p := range e.patches {
+		ps[i] = sbPatch{off: p.off, name: p.name, addend: p.addend}
+	}
+	return e.out, ps, nil
 }

 // encodeJump encodes a JMP/CALL/Jcc with a relative offset resolved from the
@@ -345,7 +388,7 @@ var spReg = Reg{idx: 4, size: 8}

 // operandFromAST converts a parsed operand into an encoder Operand, applying
 // the frame translation to FP/SP pseudo-register operands.
-func operandFromAST(op *ast.Operand, size int, fi frameInfo) (Operand, error) {
+func operandFromAST(op *ast.Operand, size int, fi frameInfo, link *linkInfo) (Operand, error) {
 	switch op.Kind {
 	case ast.OpImmediate:
 		if op.Imm.HasVal {
@@ -371,9 +414,20 @@ func operandFromAST(op *ast.Operand, size int, fi frameInfo) (Operand, error) {
 			off := fi.spAdjust + a.Sym.Offset
 			return Mem{Base: spReg, Disp: off, HasBase: true, Size: size}, nil
 		}
-		// SB (global symbol) needs a relocation — not yet supported.
+		// SB (global symbol): a static (file-local, <>) symbol becomes a
+		// RIP-relative reference resolved by the file-level layout; anything
+		// external needs object-file emission.
 		if a.Sym != nil && a.Sym.Pseudo == "SB" {
-			return nil, fmt.Errorf("SB (global symbol) operands need relocation support (pending)")
+			if !a.Sym.Static {
+				return nil, fmt.Errorf("external symbol %q needs object-file emission", a.Sym.Name)
+			}
+			if link == nil || link.symbols == nil {
+				return nil, fmt.Errorf("static symbol %q needs file-level assembly (AssembleFile)", a.Sym.Name)
+			}
+			if !link.symbols[a.Sym.Name] {
+				return nil, fmt.Errorf("undefined symbol %q", 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).