diff --git a/arch/arm64.go b/arch/arm64.go
index 24f6a85..9e73d7f 100644
--- a/arch/arm64.go
+++ b/arch/arm64.go
@@ -29,7 +29,7 @@ func arm64Registers() []Register {
 		regs = append(regs, Register{Name: name, Class: class, Desc: desc})
 	}
 
-	// General-purpose integer registers R0–R30.
+	// General-purpose integer registers R0-R30.
 	for i := 0; i <= 30; i++ {
 		add(fmt.Sprintf("R%d", i), GPR, "64-bit general-purpose register")
 	}
diff --git a/asm/arm64_encode.go b/asm/arm64_encode.go
index 41e58af..a09bf70 100644
--- a/asm/arm64_encode.go
+++ b/asm/arm64_encode.go
@@ -9,7 +9,7 @@ package asm
 // an opcode constant, and the format selects the bit layout.  The opcode
 // constants and formats are transcribed from the Go toolchain's own arm64
 // backend (cmd/internal/obj/arm64), so the emitted bytes match `go tool asm`
-// exactly — the ground-truth oracle for the verify suite.
+// exactly, the ground-truth oracle for the verify suite.
 //
 // All AArch64 instructions are 32 bits, little-endian.  The formats used here
 // (per the ARM Architecture Reference Manual):
@@ -28,7 +28,7 @@ package asm
 //	ADR/ADRP        p<<31 | 0x10<<24 | immlo<<29 | immhi<<5 | Rd
 
 // arm64RegNum returns the 5-bit register number for an AArch64 register name:
-// R0–R30 (integer), F0–F31 (floating point), and the ABI aliases the
+// R0-R30 (integer), F0-F31 (floating point), and the ABI aliases the
 // runtime's assembly uses.  Returns -1 for an unrecognised name.
 func arm64RegNum(name string) int {
 	switch name {
@@ -99,7 +99,7 @@ func arm64RegNum(name string) int {
 	case "SP":
 		return 31 // SP and ZR share encoding 31; context determines meaning
 	}
-	// F0–F31.
+	// F0-F31.
 	if len(name) >= 1 && name[0] == 'F' {
 		n := 0
 		for i := 1; i < len(name); i++ {
diff --git a/asm/elf.go b/asm/elf.go
index ef903dc..8776e09 100644
--- a/asm/elf.go
+++ b/asm/elf.go
@@ -12,7 +12,7 @@ import (
 // Image: a .text section holding the function bodies, a .data section
 // holding the GLOBL initialisers, a symbol table with one symbol per TEXT
 // and GLOBL (file-local <> symbols are STB_LOCAL, the rest STB_GLOBAL), and
-// a .rela.text relocation table — one R_X86_64_PC32 entry per static-symbol
+// a .rela.text relocation table, one R_X86_64_PC32 entry per static-symbol
 // reference, internal references resolving against the local data symbols
 // and external ones against undefined globals.  The output links with the
 // system toolchain (cc/ld) the way a hand-assembled .o would.
@@ -71,7 +71,7 @@ func (img *Image) ELFObject() ([]byte, error) {
 
 	// Build the symbol table: the null entry and the two section symbols
 	// come first, then the local symbols (static TEXT and GLOBL), then the
-	// globals (exported TEXT and GLOBL, and the undefined externals) — ELF
+	// globals (exported TEXT and GLOBL, and the undefined externals), ELF
 	// requires every local to precede every global, and sh_info records the
 	// boundary.  symIdx maps a symbol name to its index for the relocations.
 	var locals, globals []elfSym
@@ -218,7 +218,7 @@ func (img *Image) ELFObject() ([]byte, error) {
 	shstrOff := len(out)
 	out = append(out, stSections.bytes()...)
 
-	// DWARF debug sections (no relocations — the linker resolves DWARF fixups).
+	// DWARF debug sections (no relocations, the linker resolves DWARF fixups).
 	dwAlign := func(n int) {
 		for len(out)%n != 0 {
 			out = append(out, 0)
diff --git a/asm/elf_dwarf.go b/asm/elf_dwarf.go
index 9d85130..2589433 100644
--- a/asm/elf_dwarf.go
+++ b/asm/elf_dwarf.go
@@ -282,7 +282,7 @@ func dwarfBuildFrameSection(img *Image) []byte {
 	// Patch CIE length.
 	le.PutUint32(b[cieStart:], uint32(len(b)-cieStart-4))
 
-	// FDEs (Frame Description Entries) — one per function.
+	// FDEs (Frame Description Entries), one per function.
 	for _, fn := range img.Funcs {
 		fdeStart := len(b)
 		b = append(b, 0, 0, 0, 0)                // length (placeholder)
diff --git a/asm/encode.go b/asm/encode.go
index 24af92d..fcc2326 100644
--- a/asm/encode.go
+++ b/asm/encode.go
@@ -284,7 +284,7 @@ func setRM(i *instr, reg Reg, rm Operand, opSize int) error {
 }
 
 // setRMDigit fills in the ModR/M for an instruction whose reg field is an
-// opcode /digit extension (0–7), which carries none of the register REX rules.
+// opcode /digit extension (0-7), which carries none of the register REX rules.
 func setRMDigit(i *instr, digit int, rm Operand, opSize int) error {
 	return setRMReg(i, digit, false, false, rm, opSize)
 }
diff --git a/asm/evex.go b/asm/evex.go
index c866f3c..30cc2c3 100644
--- a/asm/evex.go
+++ b/asm/evex.go
@@ -9,10 +9,10 @@ import (
 )
 
 // This file implements EVEX (AVX-512) instruction encoding: the four-byte
-// EVEX prefix with 5-bit vector register fields (Z0–Z31, X/Y 16–31), the
+// EVEX prefix with 5-bit vector register fields (Z0-Z31, X/Y 16-31), the
 // compressed disp8×N displacement, and the operand shapes the go-flac
 // AVX-512 kernels use plus the common floating-point and conversion set.
-// Masking follows the Go assembler's spelling: an explicit K1–K7 operand
+// Masking follows the Go assembler's spelling: an explicit K1-K7 operand
 // anywhere among the operands (merging) plus a ".Z" mnemonic suffix for
 // zeroing.  K-register operands (mask destinations, KMOVW, KTESTW) are
 // supported too.
@@ -36,7 +36,7 @@ type evexSpec struct {
 // are taken from the Go assembler's opcode tables, which are authoritative
 // for byte-for-byte agreement.
 var evexTable = map[string]evexSpec{
-	// EVEX.128/256/512.66.0F — integer arithmetic / logic, NDS form.
+	// EVEX.128/256/512.66.0F, integer arithmetic / logic, NDS form.
 	"VPADDD":      {1, 0xFE, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPADDQ":      {1, 0xD4, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPSUBD":      {1, 0xFA, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
@@ -48,25 +48,25 @@ var evexTable = map[string]evexSpec{
 	"VPCMPEQD":    {1, 0x76, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VFMADD231PD": {2, 0xB8, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 
-	// EVEX.128/256/512.66.0F.W1 — packed double arithmetic.
+	// EVEX.128/256/512.66.0F.W1, packed double arithmetic.
 	"VADDPD": {1, 0x58, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VMULPD": {1, 0x59, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VSUBPD": {1, 0x5C, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VDIVPD": {1, 0x5E, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VMINPD": {1, 0x5D, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VMAXPD": {1, 0x5F, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.0F.W0 — packed single arithmetic.
+	// EVEX.128/256/512.0F.W0, packed single arithmetic.
 	"VADDPS": {1, 0x58, 0, 0, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VMULPS": {1, 0x59, 0, 0, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VSUBPS": {1, 0x5C, 0, 0, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VDIVPS": {1, 0x5E, 0, 0, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VMINPS": {1, 0x5D, 0, 0, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VMAXPS": {1, 0x5F, 0, 0, -1, vexNDS3, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.66.0F.W1 — packed double unpack.
+	// EVEX.128/256/512.66.0F.W1, packed double unpack.
 	"VUNPCKLPD": {1, 0x14, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VUNPCKHPD": {1, 0x15, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 
-	// EVEX.128.F2.0F.W1 — scalar double arithmetic (the packed opcodes with
+	// EVEX.128.F2.0F.W1, scalar double arithmetic (the packed opcodes with
 	// an F2 pp; the EVEX forms exist for masked and zeroing use).  The
 	// memory operand is a single double, so disp8×N = 8.
 	"VADDSD": {1, 0x58, 1, 3, -1, vexNDS3, [3]int{8, 8, 8}},
@@ -76,7 +76,7 @@ var evexTable = map[string]evexSpec{
 	"VMINSD": {1, 0x5D, 1, 3, -1, vexNDS3, [3]int{8, 8, 8}},
 	"VMAXSD": {1, 0x5F, 1, 3, -1, vexNDS3, [3]int{8, 8, 8}},
 
-	// EVEX.128.F3.0F.W0 — scalar single arithmetic (disp8×N = 4).
+	// EVEX.128.F3.0F.W0, scalar single arithmetic (disp8×N = 4).
 	"VADDSS": {1, 0x58, 0, 2, -1, vexNDS3, [3]int{4, 4, 4}},
 	"VSUBSS": {1, 0x5C, 0, 2, -1, vexNDS3, [3]int{4, 4, 4}},
 	"VMULSS": {1, 0x59, 0, 2, -1, vexNDS3, [3]int{4, 4, 4}},
@@ -84,38 +84,38 @@ var evexTable = map[string]evexSpec{
 	"VMINSS": {1, 0x5D, 0, 2, -1, vexNDS3, [3]int{4, 4, 4}},
 	"VMAXSS": {1, 0x5F, 0, 2, -1, vexNDS3, [3]int{4, 4, 4}},
 
-	// EVEX.512.66.0F3A — align (NDS + imm8).
+	// EVEX.512.66.0F3A, align (NDS + imm8).
 	"VALIGND": {3, 0x03, 0, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 
-	// EVEX.128/256/512.66.0F — immediate shift (VPSRAD /4).
+	// EVEX.128/256/512.66.0F, immediate shift (VPSRAD /4).
 	"VPSRAD": {1, 0x72, 0, 1, 4, vexShiftImm, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.66.0F.W1 — variable shift with an XMM count (VPSRAQ;
+	// EVEX.128/256/512.66.0F.W1, variable shift with an XMM count (VPSRAQ;
 	// the W bit distinguishes it from VPSRAD's E2 form).
 	"VPSRAQ": {1, 0xE2, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 
-	// EVEX.128/256/512.F3.0F.W1 — signed qword to packed double (reg=dst,
+	// EVEX.128/256/512.F3.0F.W1, signed qword to packed double (reg=dst,
 	// rm=src, no vvvv).
 	"VCVTQQ2PD": {1, 0xE6, 1, 2, -1, vexRM, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.F2.0F.W1 — duplicate the low double (reg=dst,
+	// EVEX.128/256/512.F2.0F.W1, duplicate the low double (reg=dst,
 	// rm=src, no vvvv): a 128-bit destination reads a single double from
 	// memory (disp8×8), the wider ones read the full operand.
 	"VMOVDDUP": {1, 0x12, 1, 3, -1, vexRM, [3]int{8, 32, 64}},
-	// EVEX.128/256/512.0F.W0 — signed dword to packed single (reg=dst,
-	// rm=src, no vvvv, no mandatory prefix — as in the VEX form).
+	// EVEX.128/256/512.0F.W0, signed dword to packed single (reg=dst,
+	// rm=src, no vvvv, no mandatory prefix, as in the VEX form).
 	"VCVTDQ2PS": {1, 0x5B, 0, 0, -1, vexRM, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.0F.W0 — packed single to packed double: the
+	// EVEX.128/256/512.0F.W0, packed single to packed double: the
 	// destination is twice the source width and sets the length; disp8×N
 	// follows the narrow memory source.  No F3 prefix: the Go assembler
 	// emits this instruction with pp = 00 (Intel's maps would call that
-	// undefined) and gasm reproduces the Go assembler's bytes — its machine
+	// undefined) and gasm reproduces the Go assembler's bytes, its machine
 	// code is the oracle, not the manual.
 	"VCVTPS2PD": {1, 0x5A, 0, 0, -1, vexRM, [3]int{8, 16, 32}},
-	// EVEX.128/256/512.F3.0F.W0 — signed dword to packed double (the EVEX
+	// EVEX.128/256/512.F3.0F.W0, signed dword to packed double (the EVEX
 	// form of the VEX instruction; the destination sets the length, disp8×N
 	// follows the narrow memory source).
 	"VCVTDQ2PD": {1, 0xE6, 0, 2, -1, vexRM, [3]int{8, 16, 32}},
 	// EVEX packed double → dword conversions: the source is the wide
-	// operand and the mnemonic fixes the length — the bare names are
+	// operand and the mnemonic fixes the length, the bare names are
 	// 512-bit only (ZMM source, XMM destination), the X/Y spellings are
 	// EVEX-128/256.  Exactly one slot of n is valid; it names the vector
 	// length (and the disp8×N multiplier) a register or memory source
@@ -127,7 +127,7 @@ var evexTable = map[string]evexSpec{
 	"VCVTTPD2DQX": {1, 0xE6, 1, 1, -1, vexRMSrcLen, [3]int{16, 0, 0}},
 	"VCVTTPD2DQY": {1, 0xE6, 1, 1, -1, vexRMSrcLen, [3]int{0, 32, 0}},
 
-	// EVEX.66.0F3A — ternary logic and lane shuffles (NDS + imm8).
+	// EVEX.66.0F3A, ternary logic and lane shuffles (NDS + imm8).
 	"VPTERNLOGD": {3, 0x25, 0, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 	"VPTERNLOGQ": {3, 0x25, 1, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 	"VSHUFI32X4": {3, 0x43, 0, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
@@ -136,11 +136,11 @@ var evexTable = map[string]evexSpec{
 	"VSHUFF64X2": {3, 0x23, 1, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 	"VPALIGNR":   {3, 0x0F, 0, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 
-	// EVEX.66.0F — the EVEX forms of the VEX two-source shuffle.
+	// EVEX.66.0F, the EVEX forms of the VEX two-source shuffle.
 	"VSHUFPD": {1, 0xC6, 1, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 	"VSHUFPS": {1, 0xC6, 0, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 
-	// EVEX.66.0F3A — lane insert ($imm, xsrc, zsrc1, zdst).
+	// EVEX.66.0F3A, lane insert ($imm, xsrc, zsrc1, zdst).
 	"VINSERTF32X4": {3, 0x18, 0, 1, -1, vexNDS3Imm, [3]int{0, 16, 32}},
 	"VINSERTF32X8": {3, 0x1A, 0, 1, -1, vexNDS3Imm, [3]int{0, 0, 32}},
 	"VINSERTF64X2": {3, 0x18, 1, 1, -1, vexNDS3Imm, [3]int{0, 16, 32}},
@@ -150,7 +150,7 @@ var evexTable = map[string]evexSpec{
 	"VINSERTI64X2": {3, 0x38, 1, 1, -1, vexNDS3Imm, [3]int{0, 16, 32}},
 	"VINSERTI64X4": {3, 0x3A, 1, 1, -1, vexNDS3Imm, [3]int{0, 0, 32}},
 
-	// EVEX.66.0F3A — lane extract (reg=source, rm=XMM/YMM destination,
+	// EVEX.66.0F3A, lane extract (reg=source, rm=XMM/YMM destination,
 	// imm8).
 	"VEXTRACTF32X4": {3, 0x19, 0, 1, -1, vexExtract, [3]int{0, 16, 16}},
 	"VEXTRACTF32X8": {3, 0x1B, 0, 1, -1, vexExtract, [3]int{0, 0, 32}},
@@ -159,14 +159,14 @@ var evexTable = map[string]evexSpec{
 	"VEXTRACTI32X8": {3, 0x3B, 0, 1, -1, vexExtract, [3]int{0, 0, 32}},
 	"VEXTRACTI64X2": {3, 0x39, 1, 1, -1, vexExtract, [3]int{0, 16, 16}},
 
-	// EVEX.66.0F — compare with an opmask destination ($imm, src2, src1,
+	// EVEX.66.0F, compare with an opmask destination ($imm, src2, src1,
 	// kdst): NDS3Imm with the K register in the reg field.
 	"VCMPPD": {1, 0xC2, 1, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 	"VCMPPS": {1, 0xC2, 0, 0, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 	"VCMPSD": {1, 0xC2, 1, 3, -1, vexNDS3Imm, [3]int{8, 8, 8}},
 	"VCMPSS": {1, 0xC2, 0, 2, -1, vexNDS3Imm, [3]int{4, 4, 4}},
 
-	// EVEX.66.0F3A — integer compares with an opmask destination, the same
+	// EVEX.66.0F3A, integer compares with an opmask destination, the same
 	// NDS3Imm-with-k-reg shape as the floating-point compares; W selects the
 	// operand width (byte/word vs dword/qword), the opcode the signedness.
 	// The memory form takes a full vector, so disp8×N is 16/32/64.
@@ -179,7 +179,7 @@ var evexTable = map[string]evexSpec{
 	"VPCMPQ":  {3, 0x1F, 1, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 	"VPCMPUQ": {3, 0x1E, 1, 1, -1, vexNDS3Imm, [3]int{16, 32, 64}},
 
-	// EVEX.66.0F38 — permutes (NDS form).
+	// EVEX.66.0F38, permutes (NDS form).
 	"VPERMB":    {2, 0x8D, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPERMW":    {2, 0x8D, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPERMI2D":  {2, 0x76, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
@@ -188,7 +188,7 @@ var evexTable = map[string]evexSpec{
 	"VPERMT2Q":  {2, 0x7E, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPERMT2PD": {2, 0x7F, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 
-	// EVEX.66.0F — the wider integer set (NDS form).
+	// EVEX.66.0F, the wider integer set (NDS form).
 	"VPMADDWD":   {1, 0xF5, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPMULHUW":   {1, 0xE4, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPMADDUBSW": {2, 0x04, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
@@ -199,34 +199,34 @@ var evexTable = map[string]evexSpec{
 	"VPACKSSDW":  {1, 0x6B, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPACKUSDW":  {2, 0x2B, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 
-	// EVEX.66.0F38 — absolute values and replicating moves (reg=dst,
+	// EVEX.66.0F38, absolute values and replicating moves (reg=dst,
 	// rm=src).
 	"VPABSB": {2, 0x1C, 0, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VPABSW": {2, 0x1D, 0, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VPABSD": {2, 0x1E, 0, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VPABSQ": {2, 0x1F, 1, 1, -1, vexRM, [3]int{16, 32, 64}},
-	// EVEX.F3.0F — replicate even/odd singles.
+	// EVEX.F3.0F, replicate even/odd singles.
 	"VMOVSLDUP": {1, 0x12, 0, 2, -1, vexRM, [3]int{16, 32, 64}},
 	"VMOVSHDUP": {1, 0x16, 0, 2, -1, vexRM, [3]int{16, 32, 64}},
-	// EVEX.66.0F38 — sign/zero-extending moves; the memory source is the
+	// EVEX.66.0F38, sign/zero-extending moves; the memory source is the
 	// narrow half (here byte to word).
 	"VPMOVSXBW": {2, 0x20, 0, 1, -1, vexRM, [3]int{8, 16, 32}},
 	"VPMOVZXBW": {2, 0x30, 0, 1, -1, vexRM, [3]int{8, 16, 32}},
-	// EVEX.66.0F — packed single conversions (reg=dst, rm=src).
+	// EVEX.66.0F, packed single conversions (reg=dst, rm=src).
 	"VCVTPS2DQ":  {1, 0x5B, 0, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VCVTTPS2DQ": {1, 0x5B, 0, 2, -1, vexRM, [3]int{16, 32, 64}},
-	// EVEX.66.0F38 — broadcast a single/double to all lanes (reg=dst,
+	// EVEX.66.0F38, broadcast a single/double to all lanes (reg=dst,
 	// rm=scalar memory; disp8×N is the element size).
 	"VBROADCASTSS": {2, 0x18, 0, 1, -1, vexRM, [3]int{4, 4, 4}},
 	"VBROADCASTSD": {2, 0x19, 1, 1, -1, vexRM, [3]int{0, 8, 8}},
 
-	// EVEX.66.0F38 — expand loads (rm → vector register destination).
+	// EVEX.66.0F38, expand loads (rm → vector register destination).
 	"VEXPANDPD": {2, 0x88, 1, 1, -1, vexRM, [3]int{8, 8, 8}},
 	"VEXPANDPS": {2, 0x88, 0, 1, -1, vexRM, [3]int{4, 4, 4}},
 	"VPEXPANDD": {2, 0x89, 0, 1, -1, vexRM, [3]int{4, 4, 4}},
 	"VPEXPANDQ": {2, 0x89, 1, 1, -1, vexRM, [3]int{8, 8, 8}},
 
-	// EVEX.66.0F38 — compress stores (vector register source → rm), and the
+	// EVEX.66.0F38, compress stores (vector register source → rm), and the
 	// remaining narrowing stores.
 	"VCOMPRESSPD": {2, 0x8A, 1, 1, -1, vexRMRev, [3]int{8, 8, 8}},
 	"VCOMPRESSPS": {2, 0x8A, 0, 1, -1, vexRMRev, [3]int{4, 4, 4}},
@@ -235,7 +235,7 @@ var evexTable = map[string]evexSpec{
 	"VPMOVWB":     {2, 0x30, 0, 2, -1, vexRMRev, [3]int{8, 16, 32}},
 	"VPMOVQB":     {2, 0x32, 0, 2, -1, vexRMRev, [3]int{2, 4, 8}},
 
-	// EVEX.66.0F — rotates (immediate form: /0 right, /1 left).
+	// EVEX.66.0F, rotates (immediate form: /0 right, /1 left).
 	"VPRORD": {1, 0x72, 0, 1, 0, vexShiftImm, [3]int{16, 32, 64}},
 	"VPRORQ": {1, 0x72, 1, 1, 0, vexShiftImm, [3]int{16, 32, 64}},
 	"VPROLD": {1, 0x72, 0, 1, 1, vexShiftImm, [3]int{16, 32, 64}},
@@ -248,14 +248,14 @@ var evexTable = map[string]evexSpec{
 	"VPSRLQ": {1, 0x73, 1, 1, 2, vexShiftImm, [3]int{16, 32, 64}},
 	"VPSLLQ": {1, 0x73, 1, 1, 6, vexShiftImm, [3]int{16, 32, 64}},
 
-	// EVEX.66.0F38 — floating-point helpers, packed (reg=dst, rm=src).
+	// EVEX.66.0F38, floating-point helpers, packed (reg=dst, rm=src).
 	"VRCP14PD":   {2, 0x4C, 1, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VRCP14PS":   {2, 0x4C, 0, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VRSQRT14PD": {2, 0x4E, 1, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VRSQRT14PS": {2, 0x4E, 0, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VGETEXPPD":  {2, 0x42, 1, 1, -1, vexRM, [3]int{16, 32, 64}},
 	"VGETEXPPS":  {2, 0x42, 0, 1, -1, vexRM, [3]int{16, 32, 64}},
-	// EVEX.66.0F38 — floating-point helpers, scalar (NDS form: src2 is
+	// EVEX.66.0F38, floating-point helpers, scalar (NDS form: src2 is
 	// rm, src1 is vvvv, the XMM destination is reg).  Like the scalar 0F3A
 	// forms, these take the 66 prefix; W selects double/single.
 	"VRCP14SD":   {2, 0x4D, 1, 1, -1, vexNDS3, [3]int{8, 8, 8}},
@@ -264,13 +264,13 @@ var evexTable = map[string]evexSpec{
 	"VRSQRT14SS": {2, 0x4F, 0, 1, -1, vexNDS3, [3]int{4, 4, 4}},
 	"VGETEXPSD":  {2, 0x43, 1, 1, -1, vexNDS3, [3]int{8, 8, 8}},
 	"VGETEXPSS":  {2, 0x43, 0, 1, -1, vexNDS3, [3]int{4, 4, 4}},
-	// EVEX.66.0F38 — scale by a power of two (NDS form).
+	// EVEX.66.0F38, scale by a power of two (NDS form).
 	"VSCALEFPD": {2, 0x2C, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VSCALEFPS": {2, 0x2C, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VSCALEFSD": {2, 0x2D, 1, 1, -1, vexNDS3, [3]int{8, 8, 8}},
 	"VSCALEFSS": {2, 0x2D, 0, 1, -1, vexNDS3, [3]int{4, 4, 4}},
 
-	// EVEX.66.0F3A — packed round/getmant/reduce ($imm, src, dst: reg=dst,
+	// EVEX.66.0F3A, packed round/getmant/reduce ($imm, src, dst: reg=dst,
 	// rm=src, imm8).
 	"VRNDSCALEPD": {3, 0x09, 1, 1, -1, vexImmRM, [3]int{16, 32, 64}},
 	"VRNDSCALEPS": {3, 0x08, 0, 1, -1, vexImmRM, [3]int{16, 32, 64}},
@@ -278,7 +278,7 @@ var evexTable = map[string]evexSpec{
 	"VGETMANTPS":  {3, 0x26, 0, 1, -1, vexImmRM, [3]int{16, 32, 64}},
 	"VREDUCEPD":   {3, 0x56, 1, 1, -1, vexImmRM, [3]int{16, 32, 64}},
 	"VREDUCEPS":   {3, 0x56, 0, 1, -1, vexImmRM, [3]int{16, 32, 64}},
-	// EVEX.66.0F3A — scalar round/getmant/reduce and fixup/range (NDS +
+	// EVEX.66.0F3A, scalar round/getmant/reduce and fixup/range (NDS +
 	// imm8: $imm, src2, src1, dst).  The scalar 0F3A forms all take the 66
 	// prefix; W selects double/single.
 	"VRNDSCALESD": {3, 0x0B, 1, 1, -1, vexNDS3Imm, [3]int{8, 8, 8}},
@@ -296,7 +296,7 @@ var evexTable = map[string]evexSpec{
 	"VRANGESD":    {3, 0x51, 1, 1, -1, vexNDS3Imm, [3]int{8, 8, 8}},
 	"VRANGESS":    {3, 0x51, 0, 1, -1, vexNDS3Imm, [3]int{4, 4, 4}},
 
-	// EVEX.66.0F3A — floating-point class test ($imm, src, kdst): the
+	// EVEX.66.0F3A, floating-point class test ($imm, src, kdst): the
 	// reg field carries the opmask destination.  The packed forms carry an
 	// explicit length in the mnemonic (X/Y/Z).
 	"VFPCLASSPDX": {3, 0x66, 1, 1, -1, vexImmRM, [3]int{16, 0, 0}},
@@ -308,7 +308,7 @@ var evexTable = map[string]evexSpec{
 	"VFPCLASSSD":  {3, 0x67, 1, 1, -1, vexImmRM, [3]int{8, 0, 0}},
 	"VFPCLASSSS":  {3, 0x67, 0, 1, -1, vexImmRM, [3]int{4, 0, 0}},
 
-	// EVEX — the remaining conversions.  VCVTQQ2PS narrows (the 512-bit
+	// EVEX, the remaining conversions.  VCVTQQ2PS narrows (the 512-bit
 	// source sets the length); the rest follow the destination.
 	"VCVTQQ2PS":  {1, 0x5B, 1, 0, -1, vexRMSrcLen, [3]int{0, 0, 64}},
 	"VCVTPD2QQ":  {1, 0x7B, 1, 1, -1, vexRM, [3]int{16, 32, 64}},
@@ -316,13 +316,13 @@ var evexTable = map[string]evexSpec{
 	"VCVTPS2QQ":  {1, 0x7B, 0, 1, -1, vexRM, [3]int{8, 16, 32}},
 	"VCVTUDQ2PD": {1, 0x7A, 0, 2, -1, vexRM, [3]int{8, 16, 32}},
 	"VCVTUDQ2PS": {1, 0x7A, 0, 0, -1, vexRM, [3]int{8, 16, 32}},
-	// EVEX.66.0F38 — half-precision convert (half-width source).
+	// EVEX.66.0F38, half-precision convert (half-width source).
 	"VCVTPH2PS": {2, 0x13, 0, 1, -1, vexRM, [3]int{8, 16, 32}},
-	// EVEX.66.0F3A — half-precision convert back ($imm, src, dst: reg=src,
-	// rm=dst, imm8 — the extract layout).
+	// EVEX.66.0F3A, half-precision convert back ($imm, src, dst: reg=src,
+	// rm=dst, imm8, the extract layout).
 	"VCVTPS2PH": {3, 0x1D, 0, 1, -1, vexExtract, [3]int{8, 16, 32}},
 
-	// EVEX — unsigned and truncating conversions.  The PD sources are the
+	// EVEX, unsigned and truncating conversions.  The PD sources are the
 	// wide operand (the bare names are 512-bit only, the X/Y spellings fix
 	// the length); the PS/UQQ destinations are wide and follow the
 	// destination.
@@ -349,7 +349,7 @@ var evexTable = map[string]evexSpec{
 	"VCVTQQ2PSX":   {1, 0x5B, 1, 0, -1, vexRMSrcLen, [3]int{16, 0, 0}},
 	"VCVTQQ2PSY":   {1, 0x5B, 1, 0, -1, vexRMSrcLen, [3]int{0, 32, 0}},
 
-	// EVEX.66.0F38 — the remaining sign/zero-extending moves (narrow
+	// EVEX.66.0F38, the remaining sign/zero-extending moves (narrow
 	// source; disp8×N follows its size).
 	"VPMOVSXBD": {2, 0x21, 0, 1, -1, vexRM, [3]int{4, 8, 16}},
 	"VPMOVSXBQ": {2, 0x22, 0, 1, -1, vexRM, [3]int{2, 4, 8}},
@@ -361,7 +361,7 @@ var evexTable = map[string]evexSpec{
 	"VPMOVZXWQ": {2, 0x34, 0, 1, -1, vexRM, [3]int{4, 8, 16}},
 	"VPMOVZXDQ": {2, 0x35, 0, 1, -1, vexRM, [3]int{8, 16, 32}},
 
-	// EVEX.F3.0F38 — the remaining narrowing stores (vector source in reg,
+	// EVEX.F3.0F38, the remaining narrowing stores (vector source in reg,
 	// narrow destination in r/m): signed, unsigned and the D/Q truncations.
 	"VPMOVSDB":  {2, 0x21, 0, 2, -1, vexRMRev, [3]int{4, 8, 16}},
 	"VPMOVSQB":  {2, 0x22, 0, 2, -1, vexRMRev, [3]int{2, 4, 8}},
@@ -378,7 +378,7 @@ var evexTable = map[string]evexSpec{
 	"VPMOVDB":   {2, 0x31, 0, 2, -1, vexRMRev, [3]int{4, 8, 16}},
 	"VPMOVQW":   {2, 0x34, 0, 2, -1, vexRMRev, [3]int{4, 8, 16}},
 
-	// EVEX.F3.0F38 — mask/vector conversions: M2* moves an opmask register
+	// EVEX.F3.0F38, mask/vector conversions: M2* moves an opmask register
 	// into a vector (rm = K source, reg = vector destination), *2M does the
 	// reverse (reg = K destination, rm = vector source, the length follows
 	// the vector).
@@ -391,7 +391,7 @@ var evexTable = map[string]evexSpec{
 	"VPMOVD2M": {2, 0x39, 0, 2, -1, vexRM, [3]int{16, 32, 64}},
 	"VPMOVQ2M": {2, 0x39, 1, 2, -1, vexRM, [3]int{16, 32, 64}},
 
-	// EVEX — scalar conversions between vector and general-purpose
+	// EVEX, scalar conversions between vector and general-purpose
 	// registers.  Vector to GPR (two operands: vec/mem source, GPR
 	// destination, vvvv unused): the signed and truncated pair, and the
 	// unsigned forms (EVEX only).
@@ -421,22 +421,22 @@ var evexTable = map[string]evexSpec{
 	"VCVTUSI2SDQ": {1, 0x7B, 1, 3, -1, vexNDS3, [3]int{8, 8, 8}},
 	"VCVTUSI2SSL": {1, 0x7B, 0, 2, -1, vexNDS3, [3]int{4, 4, 4}},
 	"VCVTUSI2SSQ": {1, 0x7B, 1, 2, -1, vexNDS3, [3]int{8, 8, 8}},
-	// EVEX.128/256/512.66.0F38.W0 — sign-extend dwords to qwords; the memory
+	// EVEX.128/256/512.66.0F38.W0, sign-extend dwords to qwords; the memory
 	// operand is the narrow source, so disp8×N follows its size (8/16/32 for
 	// the xmm/ymm/zmm destination lengths).
 	"VPMOVSXDQ": {2, 0x25, 0, 1, -1, vexRM, [3]int{8, 16, 32}},
 
-	// EVEX.512.66.0F3A.W1 — lane extract (reg=ZMM source, rm=YMM/memory
+	// EVEX.512.66.0F3A.W1, lane extract (reg=ZMM source, rm=YMM/memory
 	// destination, imm8).
 	"VEXTRACTI64X4": {3, 0x3B, 1, 1, -1, vexExtract, [3]int{0, 0, 32}},
 	"VEXTRACTF64X4": {3, 0x1B, 1, 1, -1, vexExtract, [3]int{0, 0, 32}},
 
-	// EVEX.66.0F38 — more integer NDS forms (W distinguishes D/Q).
+	// EVEX.66.0F38, more integer NDS forms (W distinguishes D/Q).
 	"VPMULLD": {2, 0x40, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPMULLQ": {2, 0x40, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 	"VPERMD":  {2, 0x36, 0, 1, -1, vexNDS3, [3]int{0, 32, 64}},
 
-	// EVEX.128/256/512 — the wider integer set (AVX-512 F/BW): byte/word
+	// EVEX.128/256/512, the wider integer set (AVX-512 F/BW): byte/word
 	// arithmetic, the bitwise ops with D/Q suffixes, min/max, averages and
 	// variable shifts.  All NDS form; W distinguishes element size.
 	"VPADDB":  {1, 0xFC, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
@@ -474,21 +474,21 @@ var evexTable = map[string]evexSpec{
 	"VPSRAVQ": {2, 0x46, 1, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 
 	// EVEX forms of instructions that also exist in VEX (selected when a ZMM
-	// or K register, or indices 16–31, demand EVEX).
+	// or K register, or indices 16-31, demand EVEX).
 	"VPSHUFD": {1, 0x70, 0, 1, -1, vexImmRM, [3]int{16, 32, 64}},
 	"VPSHUFB": {2, 0x00, 0, 1, -1, vexNDS3, [3]int{16, 32, 64}},
 
-	// EVEX.66.0F — immediate shift (VPSLLD /6).
+	// EVEX.66.0F, immediate shift (VPSLLD /6).
 	"VPSLLD": {1, 0x72, 0, 1, 6, vexShiftImm, [3]int{16, 32, 64}},
 
-	// EVEX.F3.0F38.W0 — narrowing stores: reg = wide source, rm = narrow
+	// EVEX.F3.0F38.W0, narrowing stores: reg = wide source, rm = narrow
 	// destination (VPMOVDW dword→word, VPMOVQD qword→dword).
 	"VPMOVDW": {2, 0x33, 0, 2, -1, vexRMRev, [3]int{8, 16, 32}},
 	"VPMOVQD": {2, 0x35, 0, 2, -1, vexRMRev, [3]int{8, 16, 32}},
 }
 
 // evexBcastSpec describes an EVEX broadcast (VPBROADCASTD/Q): the opcode
-// depends on the source kind — a GPR source uses opReg, a memory source uses
+// depends on the source kind, a GPR source uses opReg, a memory source uses
 // opMem with a disp8×N of n.
 type evexBcastSpec struct {
 	mapSel int
@@ -499,10 +499,10 @@ type evexBcastSpec struct {
 }
 
 var evexBcastTable = map[string]evexBcastSpec{
-	// EVEX.128/256/512.66.0F38 — broadcast a dword/qword to all lanes.
+	// EVEX.128/256/512.66.0F38, broadcast a dword/qword to all lanes.
 	"VPBROADCASTD": {2, 0x7C, 0x58, 0, 4},
 	"VPBROADCASTQ": {2, 0x7C, 0x59, 1, 8},
-	// EVEX.128/256/512.66.0F38 — broadcast a byte/word (GPR or memory
+	// EVEX.128/256/512.66.0F38, broadcast a byte/word (GPR or memory
 	// source) to all lanes.
 	"VPBROADCASTB": {2, 0x7A, 0x78, 0, 1},
 	"VPBROADCASTW": {2, 0x7B, 0x79, 0, 2},
@@ -521,26 +521,26 @@ type evexMoveSpec struct {
 
 // evexMoveTable maps an upper-case EVEX move mnemonic to its encoding.
 var evexMoveTable = map[string]evexMoveSpec{
-	// EVEX.128/256/512.F3.0F.W0 — unaligned integer move.
+	// EVEX.128/256/512.F3.0F.W0, unaligned integer move.
 	"VMOVDQU32": {1, 2, 0x6F, 0x7F, 0, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.F3.0F.W1 — unaligned qword move.
+	// EVEX.128/256/512.F3.0F.W1, unaligned qword move.
 	"VMOVDQU64": {1, 2, 0x6F, 0x7F, 1, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.F2.0F.W0 — unaligned byte move (byte/word moves use the
+	// EVEX.128/256/512.F2.0F.W0, unaligned byte move (byte/word moves use the
 	// F2 prefix, dword/qword moves F3; the element size only changes the tuple
 	// semantics).
 	"VMOVDQU8": {1, 3, 0x6F, 0x7F, 0, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.F2.0F.W1 — unaligned word move (shares the qword
+	// EVEX.128/256/512.F2.0F.W1, unaligned word move (shares the qword
 	// encoding).
 	"VMOVDQU16": {1, 3, 0x6F, 0x7F, 1, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.66.0F.W1 — unaligned packed double move.
+	// EVEX.128/256/512.66.0F.W1, unaligned packed double move.
 	"VMOVUPD": {1, 1, 0x10, 0x11, 1, [3]int{16, 32, 64}},
-	// EVEX.128/256/512 — aligned packed moves.
+	// EVEX.128/256/512, aligned packed moves.
 	"VMOVAPS": {1, 0, 0x28, 0x29, 0, [3]int{16, 32, 64}},
 	"VMOVAPD": {1, 1, 0x28, 0x29, 1, [3]int{16, 32, 64}},
-	// EVEX.128/256/512.66.0F — aligned integer moves.
+	// EVEX.128/256/512.66.0F, aligned integer moves.
 	"VMOVDQA32": {1, 1, 0x6F, 0x7F, 0, [3]int{16, 32, 64}},
 	"VMOVDQA64": {1, 1, 0x6F, 0x7F, 1, [3]int{16, 32, 64}},
-	// EVEX.128.F3.0F.W0 — scalar single move, memory operands (the
+	// EVEX.128.F3.0F.W0, scalar single move, memory operands (the
 	// three-operand register form is not supported).
 	"VMOVSS": {1, 2, 0x10, 0x11, 0, [3]int{4, 4, 4}},
 }
@@ -559,7 +559,7 @@ func isEvex(mnemUpper string) bool {
 
 // evexRequired reports whether the operands force the EVEX encoding of a
 // mnemonic that also has a VEX form: ZMM and K registers do, and so do
-// register indices 16–31, which only EVEX can represent (X16–Y31 exist
+// register indices 16-31, which only EVEX can represent (X16-Y31 exist
 // solely under AVX-512).
 func evexRequired(upper string, ops []Operand) bool {
 	_, inVex := vexTable[upper]
@@ -578,7 +578,7 @@ func evexRequired(upper string, ops []Operand) bool {
 // evexSuffix carries the EVEX mnemonic suffixes the Go assembler accepts:
 // zeroing (.Z), a rounding mode (.RN_SAE, .RD_SAE, .RU_SAE, .RZ_SAE),
 // suppress-all-exceptions (.SAE) and memory broadcast (.BCST).  Masking is
-// not a suffix — Go writes it as an explicit K operand.
+// not a suffix, Go writes it as an explicit K operand.
 type evexSuffix struct {
 	zeroing  bool
 	sae      bool
@@ -668,7 +668,7 @@ var evexRound = map[string]bool{
 }
 
 // evexBcstN maps an instruction accepting .BCST to the broadcast element
-// size — the disp8×N multiplier for its memory operand.
+// size, the disp8×N multiplier for its memory operand.
 var evexBcstN = map[string]int{
 	"VADDPD": 8, "VSUBPD": 8, "VMULPD": 8, "VDIVPD": 8,
 	"VMINPD": 8, "VMAXPD": 8,
@@ -689,7 +689,7 @@ var evexBcstN = map[string]int{
 	"VCVTTPD2QQ": 8, "VCVTTPS2QQ": 4, "VCVTUQQ2PD": 8, "VCVTUQQ2PS": 8,
 }
 
-// splitMask extracts an explicit mask register (K1–K7) from the operand list,
+// splitMask extracts an explicit mask register (K1-K7) from the operand list,
 // returning the remaining operands and the mask index.  K0 is not a usable
 // mask (aaa = 0 means "no mask"), matching the assembler.
 func splitMask(ops []Operand) ([]Operand, int, error) {
@@ -712,7 +712,7 @@ func splitMask(ops []Operand) ([]Operand, int, error) {
 }
 
 // encodeEvex encodes an EVEX instruction with operands in Plan 9 order.  The
-// mask, when present, is an explicit K1–K7 operand anywhere among the
+// mask, when present, is an explicit K1-K7 operand anywhere among the
 // operands; the mnemonic suffix carries zeroing, rounding/SAE and
 // broadcast.
 func (e *enc) encodeEvex(mnemUpper string, ops []Operand, sfx evexSuffix) error {
@@ -1042,7 +1042,7 @@ func (e *enc) encodeEvexMove(mnem string, ms evexMoveSpec, ops []Operand, mask i
 }
 
 // encodeEvexRMSrcLen encodes a length-narrowing conversion: OP src, dst with
-// the destination always XMM and the length fixed by the mnemonic — the
+// the destination always XMM and the length fixed by the mnemonic, the
 // single valid slot of spec.n names the vector length (and the disp8×N
 // multiplier) a register or memory source encodes.
 func (e *enc) encodeEvexRMSrcLen(spec evexSpec, ops []Operand, mask int, sfx evexSuffix) error {
@@ -1061,7 +1061,7 @@ func (e *enc) encodeEvexRMSrcLen(spec evexSpec, ops []Operand, mask int, sfx eve
 	return e.emitEvexFields(spec, ll, dstReg.idx, -1, src, mask, sfx)
 }
 
-// soleLen returns the vector-length index of the single valid slot of n —
+// soleLen returns the vector-length index of the single valid slot of n
 // the length a length-fixed mnemonic (the EVEX conversion spellings) encodes
 // regardless of its operands.
 func soleLen(n [3]int) (int, error) {
@@ -1131,8 +1131,8 @@ func (e *enc) encodeEvexBcast(bs evexBcastSpec, ops []Operand, mask int, sfx eve
 
 // emitEvexFields emits the EVEX prefix, opcode, ModR/M, SIB and displacement
 // (disp8×N compressed) for the given precomputed fields.  regIdx is the
-// unextended reg-field register index, or a /digit (0–7); vvvvIdx is the
-// vvvv register index, or -1 when unused.  mask (K1–K7, 0 = unmasked) and
+// unextended reg-field register index, or a /digit (0-7); vvvvIdx is the
+// vvvv register index, or -1 when unused.  mask (K1-K7, 0 = unmasked) and
 // zeroing fill the aaa and z bits of the P2 byte.
 func (e *enc) emitEvexFields(spec evexSpec, ll, regIdx, vvvvIdx int, rm Operand, mask int, sfx evexSuffix) error {
 	if ll > 2 {
@@ -1324,7 +1324,7 @@ func isScatter(upper string) bool {
 }
 
 // vsibLen validates a VSIB memory operand (the index must be a vector
-// register) and returns it with the vector length the index selects — the
+// register) and returns it with the vector length the index selects, the
 // EVEX L'L field follows the index register, not the data register.
 func vsibLen(op Operand, what string) (Mem, int, error) {
 	m, ok := op.(Mem)
@@ -1383,7 +1383,7 @@ func (e *enc) encodeGather(upper string, gs gatherSpec, ops []Operand, sfx evexS
 	return e.emitVexFields(spec, dst.vecLenBit(), dst.idx&7, rBit, 15-maskReg.idx, vsib)
 }
 
-// encodeScatter encodes a scatter (EVEX only): OP src, K, vsib — reg = src,
+// encodeScatter encodes a scatter (EVEX only): OP src, K, vsib, reg = src,
 // rm = the VSIB memory operand, the K mask in aaa and L following the VSIB
 // index.
 func (e *enc) encodeScatter(upper string, ss gatherSpec, ops []Operand, sfx evexSuffix) error {
@@ -1411,15 +1411,15 @@ func (e *enc) encodeScatter(upper string, ss gatherSpec, ops []Operand, sfx evex
 
 // evexKOperand lists the instructions whose K register is a genuine operand
 // (the source or destination of a mask/vector conversion) rather than a
-// mask modifier — the M2 and 2M conversions.  They take no masking.
+// mask modifier, the M2 and 2M conversions.  They take no masking.
 var evexKOperand = map[string]bool{
 	"VPMOVM2B": true, "VPMOVM2W": true, "VPMOVM2D": true, "VPMOVM2Q": true,
 	"VPMOVB2M": true, "VPMOVW2M": true, "VPMOVD2M": true, "VPMOVQ2M": true,
 }
 
 // kmovSpec describes a KMOV width: the opcode depends on the operand
-// direction — kk (k/mem → K is 90, k → k uses the same), kmem (K → mem),
-// gprk (GPR/mem → K), kgpr (K → GPR) — and the GPR forms carry a mandatory
+// direction, kk (k/mem → K is 90, k → k uses the same), kmem (K → mem),
+// gprk (GPR/mem → K), kgpr (K → GPR), and the GPR forms carry a mandatory
 // prefix and W for the wider widths.
 type kmovSpec struct {
 	kk, kmem, gprk, kgpr byte
diff --git a/asm/goobj.go b/asm/goobj.go
index e454472..4811d22 100644
--- a/asm/goobj.go
+++ b/asm/goobj.go
@@ -14,8 +14,8 @@ import (
 	"sync"
 )
 
-// This file emits GOOBJ — the Go toolchain's object format, which cmd/link
-// consumes directly — so gasm-assembled functions drop into a go build
+// This file emits GOOBJ, the Go toolchain's object format, which cmd/link
+// consumes directly, so gasm-assembled functions drop into a go build
 // without the Go assembler.  The layout follows cmd/internal/goobj: a
 // toolchain preamble ("go object ...\n!\n"), the go120ld header with its
 // block offsets, a string table, symbol definitions, the relocation /
@@ -178,7 +178,7 @@ type dwarfRelocSet struct {
 // does with its -p flag).  srcPath names the source file recorded in the
 // object's file table and line tables.  The toolchain's object preamble is
 // captured from the installed go tool asm, so the output links with the
-// toolchain it was produced on — exactly like a real assembly object.
+// toolchain it was produced on, exactly like a real assembly object.
 func (img *Image) GOObject(pkgPath, srcPath string) ([]byte, error) {
 	pre, err := toolchainObjectPreamble()
 	if err != nil {
@@ -198,7 +198,7 @@ func (img *Image) emitGOObject(pkgPath, srcPath string, pre []byte, minLC int, r
 		return nil, fmt.Errorf("GOOBJ emission requires a package path (-p)")
 	}
 
-	// The non-package definitions first — the DWARF symbols reference the
+	// The non-package definitions first, the DWARF symbols reference the
 	// functions by these indices: per function the four pc-value tables
 	// and the function itself, as cmd/asm lays them out.
 	type npSym struct {
diff --git a/asm/goobj_resolve.go b/asm/goobj_resolve.go
index 9059b83..b27f459 100644
--- a/asm/goobj_resolve.go
+++ b/asm/goobj_resolve.go
@@ -84,7 +84,7 @@ func sortedPkgRefs(refs map[string][]string) []pkgRef {
 	for pkg, syms := range refs {
 		pkgs = append(pkgs, pkgRef{pkg, syms})
 	}
-	// Simple insertion sort — the list is tiny (usually 1–3 packages).
+	// Simple insertion sort, the list is tiny (usually 1-3 packages).
 	for i := 1; i < len(pkgs); i++ {
 		for j := i; j > 0 && pkgs[j-1].path > pkgs[j].path; j-- {
 			pkgs[j-1], pkgs[j] = pkgs[j], pkgs[j-1]
diff --git a/asm/goobjloong64.go b/asm/goobjloong64.go
index 92e6a11..a3d7aa1 100644
--- a/asm/goobjloong64.go
+++ b/asm/goobjloong64.go
@@ -13,9 +13,9 @@ import (
 )
 
 // GOObjectLOONG64 emits a GOOBJ object file for LoongArch.  The layout is
-// the shared one in goobj.go — the toolchain preamble, the go120ld header
+// the shared one in goobj.go, the toolchain preamble, the go120ld header
 // with its block offsets, the string table, the symbol definitions and the
-// reloc/aux/data index arrays — with the loong64 preamble, the MinLC of 4
+// reloc/aux/data index arrays, with the loong64 preamble, the MinLC of 4
 // for the pc-value deltas, and R_LOONG64_ADDR_HI/LO relocation types for
 // the pcalau12i+addi.d address pairs.
 func (img *Image) GOObjectLOONG64(pkgPath, srcPath string) ([]byte, error) {
diff --git a/asm/goobjriscv.go b/asm/goobjriscv.go
index 4365c7f..01e0c08 100644
--- a/asm/goobjriscv.go
+++ b/asm/goobjriscv.go
@@ -13,9 +13,9 @@ import (
 )
 
 // GOObjectRISCV emits a GOOBJ object file for RISC-V.  The layout is the
-// shared one in goobj.go — the toolchain preamble, the go120ld header with
+// shared one in goobj.go, the toolchain preamble, the go120ld header with
 // its block offsets, the string table, the symbol definitions and the
-// reloc/aux/data index arrays — with the RISC-V preamble, the MinLC of 2 for
+// reloc/aux/data index arrays, with the RISC-V preamble, the MinLC of 2 for
 // the pc-value deltas, and the single R_RISCV_PCREL_ITYPE/STYPE relocation
 // per AUIPC pair, matching `go tool asm`'s model (each pair is one 8-byte
 // relocation, not the ELF HI20/LO12 pair).
diff --git a/asm/instrs.go b/asm/instrs.go
index 57bcc61..8d75d53 100644
--- a/asm/instrs.go
+++ b/asm/instrs.go
@@ -21,7 +21,7 @@ var aluOp = map[string]struct {
 }
 
 // unaryOp maps INC/DEC/NEG/NOT to their /digit and base opcode.  INC/DEC use
-// the 0xFE/0xFF group (the short 0x40–0x4F forms are REX prefixes in 64-bit
+// the 0xFE/0xFF group (the short 0x40-0x4F forms are REX prefixes in 64-bit
 // mode); NEG/NOT use the 0xF6/0xF7 group.
 var unaryOp = map[string]struct {
 	digit int
@@ -33,7 +33,7 @@ var unaryOp = map[string]struct {
 	"NEG": {3, 0xF7},
 }
 
-// shiftOp maps SHL/SHR/SAR to their /digit in the 0xC0/0xC1/0xD0–0xD3 group.
+// shiftOp maps SHL/SHR/SAR to their /digit in the 0xC0/0xC1/0xD0-0xD3 group.
 var shiftOp = map[string]int{
 	"SHL": 4,
 	"SHR": 5,
@@ -50,7 +50,7 @@ func (e *enc) encodeMov(ops []Operand, size int) error {
 
 	// Integer scalar XMM moves: MOVQ with an XMM operand is the SSE2
 	// packed-quadword move, NOT a GPR move: mem→xmm encodes as F3 0F 7E
-	// (reg = dst, no REX.W — the Go assembler's form), xmm→mem as
+	// (reg = dst, no REX.W, the Go assembler's form), xmm→mem as
 	// 66 0F D6 (rm = xmm).  Register forms against a GPR use the MOVD
 	// opcodes with REX.W instead: 66 REX.W 0F 6E (gpr→xmm) and
 	// 66 REX.W 0F 7E (xmm→gpr); the memory opcodes with a register r/m
@@ -103,7 +103,7 @@ func (e *enc) encodeMov(ops []Operand, size int) error {
 	switch src := src.(type) {
 	case Reg:
 		if dstIsReg {
-			// MOV r/m, r: 0x88/0x89, reg=src, rm=dst — the form the Go
+			// MOV r/m, r: 0x88/0x89, reg=src, rm=dst, the form the Go
 			// assembler emits for register-to-register moves.
 			i := newInstr(size, []byte{movRM(size)})
 			if err := setRM(i, src, dst, size); err != nil {
@@ -147,7 +147,7 @@ func (e *enc) encodeMov(ops []Operand, size int) error {
 			// a signed int32, choosing per sign:
 			//   v >= 0: B8+rd imm32 without REX.W (zero-extended by the
 			//           hardware, REX.B still emitted for R8-R15);
-			//   v <  0: REX.W C7 /0 imm32 (sign-extended — the plain B8+rd
+			//   v <  0: REX.W C7 /0 imm32 (sign-extended, the plain B8+rd
 			//           form would zero-extend and corrupt the value).
 			// Out-of-range immediates keep the B8+rd imm64 form.
 			if size == 8 && v >= 0 && v <= (1<<31)-1 {
@@ -226,8 +226,8 @@ func (e *enc) encodeALU(op struct {
 		return e.encodeALUImm(op.digit, dst, int64(imm), size)
 	}
 
-	// CMP accepts the immediate in the second position too — CMPL CX, $31 is
-	// the form the Go assembler itself accepts — and encodes it identically
+	// CMP accepts the immediate in the second position too, CMPL CX, $31 is
+	// the form the Go assembler itself accepts, and encodes it identically
 	// (CMP r/m, imm sets the flags as first − second).  No other ALU op takes
 	// an immediate destination.
 	if imm, ok := dst.(Imm); ok {
@@ -313,7 +313,7 @@ func (e *enc) encodeALUImm(digit int, dst Operand, imm int64, size int) error {
 		i.imm = []byte{byte(int8(imm))}
 		return e.emit(i)
 	}
-	// 0x81 /digit, imm16/imm32 — or the Go assembler's accumulator short
+	// 0x81 /digit, imm16/imm32, or the Go assembler's accumulator short
 	// form (opcode+5, no ModR/M) when the destination is AX/AL, which it
 	// prefers over the generic form exactly here.
 	if r, ok := dst.(Reg); ok && r.idx == 0 {
@@ -339,7 +339,7 @@ func (e *enc) encodeTest(ops []Operand, size int) error {
 	}
 	src, dst := ops[0], ops[1]
 	if imm, ok := src.(Imm); ok {
-		// TEST r/m, imm: 0xF6 (8-bit) / 0xF7 /0 — but the Go assembler
+		// TEST r/m, imm: 0xF6 (8-bit) / 0xF7 /0, but the Go assembler
 		// always uses the accumulator forms (A8/A9, no ModR/M) when the
 		// register operand is AL/AX, whatever the immediate's width.
 		if r, ok := dst.(Reg); ok && r.idx == 0 {
@@ -678,7 +678,7 @@ func (e *enc) encodeCmov(upper string, ops []Operand) error {
 }
 
 // encodeSet encodes a conditional byte set: SET + condition (SETNE, SETEQ, …),
-// always a byte write — 0F 90+cc /0 into a register or memory operand.
+// always a byte write, 0F 90+cc /0 into a register or memory operand.
 func (e *enc) encodeSet(upper string, ops []Operand) error {
 	if len(ops) != 1 {
 		return fmt.Errorf("SETcc expects 1 operand, got %d", len(ops))
@@ -711,8 +711,8 @@ var countOp = map[string]struct {
 	"POPCNT": {0xB8, 0xF3},
 }
 
-// encodeCount encodes the bit-scan and bit-count family — BSF (0F BC),
-// BSR (0F BD), TZCNT (F3 0F BC), LZCNT (F3 0F BD) and POPCNT (F3 0F B8) —
+// encodeCount encodes the bit-scan and bit-count family, BSF (0F BC),
+// BSR (0F BD), TZCNT (F3 0F BC), LZCNT (F3 0F BD) and POPCNT (F3 0F B8)
 // with reg = dst and rm = src.  The size suffix selects the operand width
 // (BSFQ, TZCNTL, …).  Note BSF/BSR leave the destination undefined when the
 // source is zero (unlike their F3-prefixed counterparts); callers must
@@ -801,8 +801,8 @@ type sseMove struct {
 }
 
 var sseMoveTable = map[string]sseMove{
-	"MOVOU":  {0xF3, 0x6F, 0x7F}, // MOVDQU — unaligned octa
-	"MOVO":   {0x66, 0x6F, 0x7F}, // MOVDQA — aligned octa
+	"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
diff --git a/asm/loong64_encode.go b/asm/loong64_encode.go
index c377f47..e57dd15 100644
--- a/asm/loong64_encode.go
+++ b/asm/loong64_encode.go
@@ -9,7 +9,7 @@ package asm
 // an opcode constant, and the format selects the bit layout.  The opcode
 // constants and formats are transcribed from the Go toolchain's own loong64
 // backend (cmd/internal/obj/loong64), so the emitted bytes match `go tool asm`
-// exactly — the ground-truth oracle for the verify suite.
+// exactly, the ground-truth oracle for the verify suite.
 //
 // All LoongArch instructions are 32 bits, little-endian.  The formats used
 // here (per the LoongArch Volume I specification):
@@ -33,8 +33,8 @@ package asm
 import "maps"
 
 // loong64RegNum returns the 5-bit register number for a LoongArch register
-// name: R0–R31 (integer), F0–F31 (floating point), FCC0–FCC7 (condition
-// flags), FCSR0–FCSR31 (control/status) and the ABI aliases the runtime's
+// name: R0-R31 (integer), F0-F31 (floating point), FCC0-FCC7 (condition
+// flags), FCSR0-FCSR31 (control/status) and the ABI aliases the runtime's
 // assembly uses.  Returns -1 for an unrecognised name.
 func loong64RegNum(name string) int {
 	switch name {
@@ -103,7 +103,7 @@ func loong64RegNum(name string) int {
 	case "R31", "S8":
 		return 31
 	}
-	// F0–F31, FCC0–FCC7, FCSR0–FCSR31.
+	// F0-F31, FCC0-FCC7, FCSR0-FCSR31.
 	if len(name) >= 4 && name[:4] == "FCSR" {
 		return loong64RegSpecial(name[4:], 31)
 	}
@@ -199,7 +199,7 @@ func l64rrrr(op uint32, r1, r2, r3, r4 int) uint32 {
 }
 
 // l64irir encodes a BSTRINS/BSTRPICK instruction: op | msb<<16 | rj<<5 | lsb<<10 | rd.
-// The msb/lsb fields are 6 bits wide (0–63) and are validated by the caller.
+// The msb/lsb fields are 6 bits wide (0-63) and are validated by the caller.
 func l64irir(op uint32, msb, rj, lsb, rd int) uint32 {
 	return op | uint32(msb)<<16 | uint32(rj&0x1f)<<5 | uint32(lsb)<<10 | uint32(rd&0x1f)
 }
@@ -280,7 +280,7 @@ var l64DualTable = map[string]l64DualEnc{}
 var l64InstrTable = map[string]l64Enc{}
 
 func init() {
-	// 3R — integer.
+	// 3R, integer.
 	rrr := map[string]uint32{
 		"ADD": 0x20 << 15, "ADDW": 0x20 << 15, "ADDV": 0x21 << 15, "ADDVU": 0x21 << 15,
 		"SUB": 0x22 << 15, "SUBW": 0x22 << 15, "SUBV": 0x23 << 15, "SUBVU": 0x23 << 15,
@@ -300,7 +300,7 @@ func init() {
 		"CRCWBW": 0x48 << 15, "CRCWHW": 0x49 << 15, "CRCWWW": 0x4a << 15, "CRCWVW": 0x4b << 15,
 		"CRCCWBW": 0x4c << 15, "CRCCWHW": 0x4d << 15, "CRCCWWW": 0x4e << 15, "CRCCWVW": 0x4f << 15,
 	}
-	// 3R — floating point.
+	// 3R, floating point.
 	rrr["MULF"] = 0x209 << 15
 	rrr["MULD"] = 0x20a << 15
 	rrr["DIVF"] = 0x20d << 15
@@ -390,12 +390,12 @@ func init() {
 		"ROTRV": {rrr: 0x37 << 15, imm: 0x004d << 16, shift: true},
 	})
 
-	// 2RI12 — pure immediate arithmetic (LU52ID has no register form).
+	// 2RI12, pure immediate arithmetic (LU52ID has no register form).
 	l64InstrTable["LU52ID"] = l64Enc{format: l64Firr, op: 0x00c << 22}
 	// ADDV16 (addu16i.d): 2RI16 with the immediate shifted right by 16.
 	l64InstrTable["ADDV16"] = l64Enc{format: l64Firr16, op: 0x4 << 26}
 
-	// 2RI14 — LL/SC are aliased by the Go assembler to the pointer loads and
+	// 2RI14, LL/SC are aliased by the Go assembler to the pointer loads and
 	// stores (ldptr/stptr), with the offset scaled by 4.
 	l64InstrTable["MOVWP"] = l64Enc{format: l64Firr14, op: 0x25 << 24} // stptr.w
 	l64InstrTable["MOVVP"] = l64Enc{format: l64Firr14, op: 0x27 << 24} // stptr.d
@@ -414,7 +414,7 @@ func init() {
 	// LUI is the Plan 9 spelling of lu12i.w.
 	l64InstrTable["LUI"] = l64Enc{format: l64Fir20, op: 0x0a << 25}
 
-	// 4R — fused multiply-add.
+	// 4R, fused multiply-add.
 	rrrr := map[string]uint32{
 		"FMADDF": 0x81 << 20, "FMADDD": 0x82 << 20,
 		"FMSUBF": 0x85 << 20, "FMSUBD": 0x86 << 20,
@@ -425,7 +425,7 @@ func init() {
 		l64InstrTable[m] = l64Enc{format: l64Frrrr, op: op}
 	}
 
-	// IRIR — bit-field insert/extract.
+	// IRIR, bit-field insert/extract.
 	irir := map[string]uint32{
 		"BSTRINSW":  0x3<<21 | 0x0<<15,
 		"BSTRINSV":  0x2 << 22,
@@ -436,7 +436,7 @@ func init() {
 		l64InstrTable[m] = l64Enc{format: l64Firir, op: op}
 	}
 
-	// 3RI2 — ALSL.
+	// 3RI2, ALSL.
 	irrr := map[string]uint32{
 		"ALSLW": 0x2 << 17, "ALSLWU": 0x3 << 17, "ALSLV": 0x16 << 17,
 	}
@@ -452,7 +452,7 @@ func init() {
 	// PRELD.
 	l64InstrTable["PRELD"] = l64Enc{format: l64Fpreld, op: 0x0ab << 22}
 
-	// Atomics — 3R with the AM field order (rk=value, rj=address, rd=result).
+	// Atomics, 3R with the AM field order (rk=value, rj=address, rd=result).
 	am := map[string]uint32{
 		"AMSWAPB": 0x070B8 << 15, "AMSWAPH": 0x070B9 << 15,
 		"AMSWAPW": 0x070C0 << 15, "AMSWAPV": 0x070C1 << 15,
@@ -477,7 +477,7 @@ func init() {
 }
 
 // l64FpMovTable maps (mnemonic, from-class, to-class) to the 2R opcode of the
-// register move between the integer and floating-point register banks — the
+// register move between the integer and floating-point register banks, the
 // MOVW/MOVV specials the Go assembler accepts.
 var l64FpMovTable = map[string]uint32{
 	"MOVV.R.F":    0x452a << 10, // movgr2fr.d
diff --git a/asm/loong64_frame.go b/asm/loong64_frame.go
index 2d06d63..5d0b80c 100644
--- a/asm/loong64_frame.go
+++ b/asm/loong64_frame.go
@@ -98,13 +98,13 @@ func loong64Return(fi loong64FrameInfo) []byte {
 	var ws []uint32
 	if fi.autosize != 0 {
 		if !fi.leaf {
-			// MOVV 0(R3), R1 — restore the link register.
+			// MOVV 0(R3), R1, restore the link register.
 			ws = append(ws, l64irr(l64loadStoreTable["MOVV"].ld, 0, 3, 1))
 		}
-		// ADDV $autosize, R3 — close the frame.
+		// ADDV $autosize, R3, close the frame.
 		ws = append(ws, l64irr(l64DualTable["ADDV"].imm, fi.autosize, 3, 3))
 	}
-	// jirl r0, r1, 0 — return.
+	// jirl r0, r1, 0, return.
 	ws = append(ws, l64irr16(l64branchTable["JIRL"], 0, 1, 0))
 	return l64WordsLE(ws...)
 }
diff --git a/asm/reg.go b/asm/reg.go
index 5016aa7..2fa9c27 100644
--- a/asm/reg.go
+++ b/asm/reg.go
@@ -12,33 +12,33 @@ import "maps"
 import "strings"
 
 // Reg is an x86-64 register.  In Plan 9 assembly the classic names (AX, BX, …)
-// are size-agnostic — the instruction suffix (MOVQ vs MOVL) fixes the width —
+// are size-agnostic, the instruction suffix (MOVQ vs MOVL) fixes the width
 // so the encoder keys off the register's index and lets the mnemonic supply the
 // size.  The high flag marks the legacy high-byte registers AH/CH/DH/BH, which
-// occupy indices 4–7 yet take no REX prefix, unlike SPL/BPL/SIL/DIL that share
+// occupy indices 4-7 yet take no REX prefix, unlike SPL/BPL/SIL/DIL that share
 // those indices but require one.  The mask flag marks the AVX-512 opmask
-// registers K0–K7.
+// registers K0-K7.
 type Reg struct {
 	idx  int
 	size int  // informational width implied by the name; the mnemonic decides
 	high bool // AH/CH/DH/BH
-	mask bool // K0–K7 opmask register
+	mask bool // K0-K7 opmask register
 }
 
-// Index returns the register number (0–15 for GPRs, 0–31 for vectors).
+// Index returns the register number (0-15 for GPRs, 0-31 for vectors).
 func (r Reg) Index() int { return r.idx }
 
 // Size returns the width in bytes implied by the register's name.
 func (r Reg) Size() int { return r.size }
 
-// IsMask reports whether r is an AVX-512 opmask register (K0–K7).
+// IsMask reports whether r is an AVX-512 opmask register (K0-K7).
 func (r Reg) IsMask() bool { return r.mask }
 
 func (r Reg) isOperand() {}
 
 // needsREX reports whether this register forces a REX prefix at the given
-// operand size: the extended registers R8–R15 always do, and at byte size the
-// low registers SPL/BPL/SIL/DIL (indices 4–7, not high) do as well.
+// operand size: the extended registers R8-R15 always do, and at byte size the
+// low registers SPL/BPL/SIL/DIL (indices 4-7, not high) do as well.
 func (r Reg) needsREX(opSize int) bool {
 	if r.idx >= 8 {
 		return true
@@ -133,7 +133,7 @@ func buildRegByName() map[string]Reg {
 	}
 
 	// Vector: X0..X31 (128-bit, size 16), Y0..Y31 (256-bit, size 32),
-	// Z0..Z31 (512-bit, size 64).  Indices 16–31 are only encodable in EVEX
+	// Z0..Z31 (512-bit, size 64).  Indices 16-31 are only encodable in EVEX
 	// (AVX-512) instructions; the encoder validates that through its tables.
 	for i := 0; i <= 31; i++ {
 		m["X"+itoa(i)] = Reg{idx: i, size: 16}
diff --git a/asm/riscv_assemble.go b/asm/riscv_assemble.go
index 1dd329f..99074bc 100644
--- a/asm/riscv_assemble.go
+++ b/asm/riscv_assemble.go
@@ -192,7 +192,7 @@ func encodeRISCVInstr(instr *ast.Instr, pc int, offsets map[string]int, fi riscv
 		if relocs != nil {
 			*relocs = append(*relocs, Reloc{Off: 0, After: 4, Name: op.Addr.Sym.Name, Kind: RelRISCVJal, Addend: op.Addr.Sym.Offset})
 		}
-		word = riscvJType(1, 0) // JAL X1, 0 — the linker fills the offset
+		word = riscvJType(1, 0) // JAL X1, 0, the linker fills the offset
 		return []byte{byte(word), byte(word >> 8), byte(word >> 16), byte(word >> 24)}, nil
 	case "JMP":
 		// JMP = JAL X0, target.  The Go assembler never compresses this to
@@ -395,7 +395,7 @@ func encodeRISCVInstr(instr *ast.Instr, pc int, offsets map[string]int, fi riscv
 		}
 		word = riscvSType(enc, rs1, rs2, imm)
 
-	// LR (load-reserved): INSTR (addr), dst — 2 operands.
+	// LR (load-reserved): INSTR (addr), dst, 2 operands.
 	case len(ops) == 2 && isLRInstr(mnem):
 		rs1, _ := memFromOperandWithFrame(ops[0], fi)
 		rd := regFromOperand(ops[1])
@@ -404,7 +404,7 @@ func encodeRISCVInstr(instr *ast.Instr, pc int, offsets map[string]int, fi riscv
 		}
 		word = riscvAMOType(enc, rd, rs1, 0) // rs2=0 for LR
 
-	// SC (store-conditional): INSTR src, (addr), dst — 3 operands.
+	// SC (store-conditional): INSTR src, (addr), dst, 3 operands.
 	case len(ops) == 3 && isSCInstr(mnem):
 		rs2 := regFromOperand(ops[0])
 		rs1, _ := memFromOperandWithFrame(ops[1], fi)
@@ -443,7 +443,7 @@ func encodeRISCVInstr(instr *ast.Instr, pc int, offsets map[string]int, fi riscv
 		}
 		return encodeRISCVItypeImmediate(mnem, enc, rd, rd, imm)
 
-	// Loads: rd, offset(rs1) — Plan 9 order is LD src, dst.
+	// Loads: rd, offset(rs1), Plan 9 order is LD src, dst.
 	case len(ops) == 2 && isLoadInstr(mnem):
 		rd := regFromOperand(ops[1])                    // destination (last operand)
 		rs1, imm := memFromOperandWithFrame(ops[0], fi) // memory source (first operand)
@@ -538,7 +538,7 @@ func encodeRISCVMov(instr *ast.Instr, fi riscvFrameInfo, relocs *[]Reloc) ([]byt
 
 	// Immediate → register.
 	if isImmOperand(src) {
-		// MOV $sym(SB), rd — load address of a static symbol or external.
+		// MOV $sym(SB), rd, load address of a static symbol or external.
 		if src.Imm.Sym != nil && src.Imm.Sym.Pseudo == "SB" {
 			rd := regFromOperand(dst)
 			if rd < 0 {
@@ -546,7 +546,7 @@ func encodeRISCVMov(instr *ast.Instr, fi riscvFrameInfo, relocs *[]Reloc) ([]byt
 			}
 			return encodeRISCVSBAddr(src.Imm.Sym, rd, relocs), nil
 		}
-		// MOV $sym(FP/SP), rd — not supported: immediate symbol references
+		// MOV $sym(FP/SP), rd, not supported: immediate symbol references
 		// other than SB cannot be encoded as a simple immediate.
 		if src.Imm.Sym != nil && src.Imm.Sym.Pseudo != "" {
 			return nil, fmt.Errorf("MOV $%s(%s): unsupported immediate symbol reference (only SB is supported)", src.Imm.Sym.Name, src.Imm.Sym.Pseudo)
@@ -562,7 +562,7 @@ func encodeRISCVMov(instr *ast.Instr, fi riscvFrameInfo, relocs *[]Reloc) ([]byt
 	// Memory → register (load).
 	if isMemOperand(src) && !isMemOperand(dst) {
 		rd := regFromOperand(dst)
-		// MOV sym(SB), rd — load from static data.
+		// MOV sym(SB), rd, load from static data.
 		if src.Addr.Sym != nil && src.Addr.Sym.Pseudo == "SB" {
 			if rd < 0 {
 				return nil, fmt.Errorf("MOV sym(SB): invalid destination register")
@@ -580,7 +580,7 @@ func encodeRISCVMov(instr *ast.Instr, fi riscvFrameInfo, relocs *[]Reloc) ([]byt
 	// Register → memory (store).
 	if !isMemOperand(src) && isMemOperand(dst) {
 		rs2 := regFromOperand(src)
-		// MOV rd, sym(SB) — store to static data.
+		// MOV rd, sym(SB), store to static data.
 		if dst.Addr.Sym != nil && dst.Addr.Sym.Pseudo == "SB" {
 			if rs2 < 0 {
 				return nil, fmt.Errorf("MOV rd, sym(SB): invalid source register")
@@ -1006,7 +1006,7 @@ func tryCompressRVC(instr *ast.Instr, fi riscvFrameInfo) (uint16, bool) {
 		}
 
 	case "ADDW", "SUBW":
-		// C.ADDW (0x27,1) / C.SUBW (0x27,0) — CA-type, prime regs.
+		// C.ADDW (0x27,1) / C.SUBW (0x27,0), CA-type, prime regs.
 		if len(ops) == 3 {
 			funct2 := uint32(0x0)
 			if mnem == "ADDW" {
@@ -1314,7 +1314,7 @@ func suggestLabel(target string, offsets map[string]int) string {
 	}
 	// Only suggest if the distance is small enough.
 	if bestDist <= 3 && bestDist < len(target)/2+1 {
-		return fmt.Sprintf(" — did you mean %q?", best)
+		return fmt.Sprintf("; did you mean %q?", best)
 	}
 	return ""
 }
diff --git a/asm/riscv_encode.go b/asm/riscv_encode.go
index 87de26a..d794b8a 100644
--- a/asm/riscv_encode.go
+++ b/asm/riscv_encode.go
@@ -154,7 +154,7 @@ type riscvEnc struct {
 
 // riscvInstrTable maps RISC-V mnemonics to their encoding.
 var riscvInstrTable = map[string]riscvEnc{
-	// RV64I — R-type arithmetic/logic.
+	// RV64I, R-type arithmetic/logic.
 	"ADD":  {0x33, 0x0, 0x00},
 	"SUB":  {0x33, 0x0, 0x20},
 	"SLL":  {0x33, 0x1, 0x00},
@@ -165,20 +165,20 @@ var riscvInstrTable = map[string]riscvEnc{
 	"SRA":  {0x33, 0x5, 0x20},
 	"OR":   {0x33, 0x6, 0x00},
 	"AND":  {0x33, 0x7, 0x00},
-	// RV64I — 32-bit variants (W suffix).
+	// RV64I, 32-bit variants (W suffix).
 	"ADDW": {0x3B, 0x0, 0x00},
 	"SUBW": {0x3B, 0x0, 0x20},
 	"SLLW": {0x3B, 0x1, 0x00},
 	"SRLW": {0x3B, 0x5, 0x00},
 	"SRAW": {0x3B, 0x5, 0x20},
-	// RV64I — I-type shift-immediate (shamt in rs2 field).
+	// RV64I, I-type shift-immediate (shamt in rs2 field).
 	"SLLI":  {0x13, 0x1, 0x00},
 	"SRLI":  {0x13, 0x5, 0x00},
 	"SRAI":  {0x13, 0x5, 0x20},
 	"SLLIW": {0x1B, 0x1, 0x00},
 	"SRLIW": {0x1B, 0x5, 0x00},
 	"SRAIW": {0x1B, 0x5, 0x20},
-	// RV64M — multiply/divide.
+	// RV64M, multiply/divide.
 	"MUL":    {0x33, 0x0, 0x01},
 	"MULH":   {0x33, 0x1, 0x01},
 	"MULHSU": {0x33, 0x2, 0x01},
@@ -187,13 +187,13 @@ var riscvInstrTable = map[string]riscvEnc{
 	"DIVU":   {0x33, 0x5, 0x01},
 	"REM":    {0x33, 0x6, 0x01},
 	"REMU":   {0x33, 0x7, 0x01},
-	// RV64M — 32-bit variants.
+	// RV64M, 32-bit variants.
 	"MULW":  {0x3B, 0x0, 0x01},
 	"DIVW":  {0x3B, 0x4, 0x01},
 	"DIVUW": {0x3B, 0x5, 0x01},
 	"REMW":  {0x3B, 0x6, 0x01},
 	"REMUW": {0x3B, 0x7, 0x01},
-	// RV64I — I-type arithmetic.
+	// RV64I, I-type arithmetic.
 	"ADDI":  {0x13, 0x0, 0x00},
 	"ADDIW": {0x1B, 0x0, 0x00},
 	"SLTI":  {0x13, 0x2, 0x00},
@@ -228,10 +228,10 @@ var riscvInstrTable = map[string]riscvEnc{
 	"ECALL":  {0x73, 0x0, 0x00},
 	"EBREAK": {0x73, 0x0, 0x00},
 	"FENCE":  {0x0F, 0x0, 0x00},
-	// JALR — indirect jump/call (I-type).
+	// JALR, indirect jump/call (I-type).
 	"JALR": {0x67, 0x0, 0x00},
 
-	// RV64A — atomics (AMO opcode 0x2F).
+	// RV64A, atomics (AMO opcode 0x2F).
 	// funct3: 0x2 = word, 0x3 = doubleword. funct5 in bits [31:27].
 	"AMOSWAPW": {0x2F, 0x2, 0x01 << 2},
 	"AMOSWAPD": {0x2F, 0x3, 0x01 << 2},
@@ -252,7 +252,7 @@ var riscvInstrTable = map[string]riscvEnc{
 	"AMOMINUW": {0x2F, 0x2, 0x18 << 2},
 	"AMOMINUD": {0x2F, 0x3, 0x18 << 2},
 
-	// RV64F/D — floating-point arithmetic.
+	// RV64F/D, floating-point arithmetic.
 	"FADDS":  {0x53, 0x0, 0x00},
 	"FSUBS":  {0x53, 0x0, 0x04},
 	"FMULS":  {0x53, 0x0, 0x08},
@@ -274,13 +274,13 @@ var riscvInstrTable = map[string]riscvEnc{
 	"FMIND": {0x53, 0x0, 0x15},
 	"FMAXD": {0x53, 0x1, 0x15},
 
-	// RV64A — load-reserved / store-conditional (funct5 0x02 / 0x03).
+	// RV64A, load-reserved / store-conditional (funct5 0x02 / 0x03).
 	"LRW": {0x2F, 0x2, 0x02 << 2},
 	"LRD": {0x2F, 0x3, 0x02 << 2},
 	"SCW": {0x2F, 0x2, 0x03 << 2},
 	"SCD": {0x2F, 0x3, 0x03 << 2},
 
-	// FP compare — result in integer register (funct7 0x50/0x51).
+	// FP compare, result in integer register (funct7 0x50/0x51).
 	"FEQS": {0x53, 0x2, 0x50},
 	"FLTS": {0x53, 0x1, 0x50},
 	"FLES": {0x53, 0x0, 0x50},
@@ -442,10 +442,10 @@ func riscvJType(rd int, offset int32) uint32 {
 // ---- RVC (compressed) encoding helpers ----
 
 // isRVCIntReg reports whether a register number can be encoded in the 3-bit
-// prime register field used by compressed instructions (x8–x15).
+// prime register field used by compressed instructions (x8-x15).
 func isRVCIntReg(r int) bool { return r >= 8 && r <= 15 }
 
-// rvcReg3 returns the 3-bit encoding for registers x8–x15 (0–7).
+// rvcReg3 returns the 3-bit encoding for registers x8-x15 (0-7).
 func rvcReg3(r int) uint32 { return uint32(r - 8) }
 
 // rvcCR encodes a CR-type (register) compressed instruction.
@@ -455,7 +455,7 @@ func rvcCR(funct4, rd, rs2 uint32) uint16 {
 }
 
 // rvcCI encodes a CI-type (immediate) compressed instruction.
-// Used for C.ADDI, C.LI, C.LUI, C.ADDIW — linear 6-bit immediate.
+// Used for C.ADDI, C.LI, C.LUI, C.ADDIW, linear 6-bit immediate.
 func rvcCI(funct3, rd uint32, imm uint32) uint16 {
 	return uint16((funct3 << 13) | ((imm>>5)&1)<<12 | (rd << 7) | (imm&0x1F)<<2 | 0x1)
 }
diff --git a/asm/riscv_frame.go b/asm/riscv_frame.go
index cbb9fff..e6ade22 100644
--- a/asm/riscv_frame.go
+++ b/asm/riscv_frame.go
@@ -58,12 +58,12 @@ func riscvIsLeaf(t *ast.Text) bool {
 		case "CALL":
 			return false
 		case "JAL":
-			// JAL rd, target — a call only when rd is the link register.
+			// JAL rd, target, a call only when rd is the link register.
 			if len(in.Operands) >= 2 && regFromOperand(in.Operands[0]) == 1 {
 				return false
 			}
 		case "JALR":
-			// JALR rs1, rd — a call when rd is X1; JALR offset(rs1) always
+			// JALR rs1, rd, a call when rd is X1; JALR offset(rs1) always
 			// links to X1.
 			if len(in.Operands) == 1 {
 				return false
@@ -84,12 +84,12 @@ func riscvPrologue(fi riscvFrameInfo) []byte {
 		return nil
 	}
 	var out []byte
-	// MOV LR, -autosize(SP) — SD X1, -autosize(X2).  The negative offset is
+	// MOV LR, -autosize(SP), SD X1, -autosize(X2).  The negative offset is
 	// not compressible to C.SDSP (unsigned), so it stays 4 bytes.
 	out = append(out, wordLE(riscvSType(riscvEnc{0x23, 0x3, 0x00}, 2, 1, int32(-fi.autosize)))...)
-	// ADDI $-autosize, SP, SP — open the frame (C.ADDI when it fits).
+	// ADDI $-autosize, SP, SP, open the frame (C.ADDI when it fits).
 	out = append(out, riscvSPAdjust(int32(-fi.autosize))...)
-	// MOV LR, 0(SP) — SD X1, 0(X2) → C.SDSP X1, 0.
+	// MOV LR, 0(SP), SD X1, 0(X2) → C.SDSP X1, 0.
 	c := rvcSSP(0x7, 1, 0)
 	out = append(out, byte(c), byte(c>>8))
 	return out
@@ -101,10 +101,10 @@ func riscvPrologue(fi riscvFrameInfo) []byte {
 func riscvReturn(fi riscvFrameInfo) []byte {
 	var out []byte
 	if fi.autosize != 0 {
-		// MOV 0(SP), LR — LD X1, 0(X2) → C.LDSP X1, 0.
+		// MOV 0(SP), LR, LD X1, 0(X2) → C.LDSP X1, 0.
 		c := rvcLSP(0x3, 1, 0)
 		out = append(out, byte(c), byte(c>>8))
-		// ADDI $autosize, SP, SP — close the frame (C.ADDI when it fits).
+		// ADDI $autosize, SP, SP, close the frame (C.ADDI when it fits).
 		out = append(out, riscvSPAdjust(int32(fi.autosize))...)
 	}
 	// JALR X0, 0(X1).
@@ -143,7 +143,7 @@ func riscvPrologueSpadjPC(fi riscvFrameInfo) int {
 }
 
 // riscvReturnEpilogueLen returns the byte length of the RET's epilogue up to
-// (but not including) the final JALR — the point where SP is restored.
+// (but not including) the final JALR, the point where SP is restored.
 func riscvReturnEpilogueLen(fi riscvFrameInfo) int {
 	if fi.autosize == 0 {
 		return 0
diff --git a/asm/vex.go b/asm/vex.go
index 009f99c..d58b6dc 100644
--- a/asm/vex.go
+++ b/asm/vex.go
@@ -36,11 +36,11 @@ const (
 	vexNDS3Imm
 	// vexExtract is the lane-extract form `OP $imm, ysrc, xdst`: ModRM.reg =
 	// ysrc (op1), ModRM.rm = xdst or memory (op2), imm8 = op0.  The YMM
-	// source lives in the reg field, the destination in r/m — the PEXTR-style
+	// source lives in the reg field, the destination in r/m, the PEXTR-style
 	// layout.  VEXTRACTI128 and VEXTRACTF128 use this shape.
 	vexExtract
 	// vexRMRev is the reversed two-operand form `OP src, dst` with the source
-	// in ModRM.reg and the destination in r/m — the layout of the EVEX
+	// in ModRM.reg and the destination in r/m, the layout of the EVEX
 	// narrowing stores (VPMOVDW, VPMOVQD).
 	vexRMRev
 	// vexRMSrcLen is the two-operand conversion form `OP src, dst` whose
@@ -68,7 +68,7 @@ type vexSpec struct {
 // incrementally; every entry is covered by a byte-for-byte ground-truth test
 // against the Go assembler.
 var vexTable = map[string]vexSpec{
-	// VEX.128/256.66.0F.WIG — integer arithmetic / logic / compare.
+	// VEX.128/256.66.0F.WIG, integer arithmetic / logic / compare.
 	"VPADDD":      {1, 0xFE, 0, 1, -1, vexNDS3},
 	"VPADDQ":      {1, 0xD4, 0, 1, -1, vexNDS3},
 	"VPSUBD":      {1, 0xFA, 0, 1, -1, vexNDS3},
@@ -82,7 +82,7 @@ var vexTable = map[string]vexSpec{
 	"VPUNPCKHDQ":  {1, 0x6A, 0, 1, -1, vexNDS3},
 	"VPUNPCKLQDQ": {1, 0x6C, 0, 1, -1, vexNDS3},
 	"VPACKSSDW":   {1, 0x6B, 0, 1, -1, vexNDS3},
-	// VEX.256.66.0F38.W0 — dword permute (three-operand NDS form).
+	// VEX.256.66.0F38.W0, dword permute (three-operand NDS form).
 	"VPERMD": {2, 0x36, 0, 1, -1, vexNDS3},
 	// VEX.128/256.66.0F38.WIG.
 	"VPMULLD":  {2, 0x40, 0, 1, -1, vexNDS3},
@@ -90,14 +90,14 @@ var vexTable = map[string]vexSpec{
 	"VPSHUFB":  {2, 0x00, 0, 1, -1, vexNDS3},
 	"VPCMPGTQ": {2, 0x37, 0, 1, -1, vexNDS3},
 
-	// VEX.128/256.66.0F.WIG — packed double-precision arithmetic / logic.
+	// VEX.128/256.66.0F.WIG, packed double-precision arithmetic / logic.
 	"VADDPD": {1, 0x58, 0, 1, -1, vexNDS3},
 	"VMULPD": {1, 0x59, 0, 1, -1, vexNDS3},
 	"VSUBPD": {1, 0x5C, 0, 1, -1, vexNDS3},
 	"VDIVPD": {1, 0x5E, 0, 1, -1, vexNDS3},
 	"VMINPD": {1, 0x5D, 0, 1, -1, vexNDS3},
 	"VMAXPD": {1, 0x5F, 0, 1, -1, vexNDS3},
-	// VEX.128/256.0F.WIG — packed single-precision arithmetic.
+	// VEX.128/256.0F.WIG, packed single-precision arithmetic.
 	"VADDPS":    {1, 0x58, 0, 0, -1, vexNDS3},
 	"VMULPS":    {1, 0x59, 0, 0, -1, vexNDS3},
 	"VSUBPS":    {1, 0x5C, 0, 0, -1, vexNDS3},
@@ -107,7 +107,7 @@ var vexTable = map[string]vexSpec{
 	"VXORPD":    {1, 0x57, 0, 1, -1, vexNDS3},
 	"VUNPCKHPD": {1, 0x15, 0, 1, -1, vexNDS3},
 	"VUNPCKLPD": {1, 0x14, 0, 1, -1, vexNDS3},
-	// VEX.128.F2.0F.WIG — scalar double-precision arithmetic (the packed
+	// VEX.128.F2.0F.WIG, scalar double-precision arithmetic (the packed
 	// opcodes with an F2 pp).
 	"VADDSD": {1, 0x58, 0, 3, -1, vexNDS3},
 	"VSUBSD": {1, 0x5C, 0, 3, -1, vexNDS3},
@@ -115,7 +115,7 @@ var vexTable = map[string]vexSpec{
 	"VDIVSD": {1, 0x5E, 0, 3, -1, vexNDS3},
 	"VMINSD": {1, 0x5D, 0, 3, -1, vexNDS3},
 	"VMAXSD": {1, 0x5F, 0, 3, -1, vexNDS3},
-	// VEX.128.F3.0F.WIG — scalar single-precision arithmetic (the packed
+	// VEX.128.F3.0F.WIG, scalar single-precision arithmetic (the packed
 	// opcodes with an F3 pp).
 	"VADDSS": {1, 0x58, 0, 2, -1, vexNDS3},
 	"VSUBSS": {1, 0x5C, 0, 2, -1, vexNDS3},
@@ -123,10 +123,10 @@ var vexTable = map[string]vexSpec{
 	"VDIVSS": {1, 0x5E, 0, 2, -1, vexNDS3},
 	"VMINSS": {1, 0x5D, 0, 2, -1, vexNDS3},
 	"VMAXSS": {1, 0x5F, 0, 2, -1, vexNDS3},
-	// VEX.128/256.66.0F38.W1 — fused multiply-add (NDS form).
+	// VEX.128/256.66.0F38.W1, fused multiply-add (NDS form).
 	"VFMADD231PD": {2, 0xB8, 1, 1, -1, vexNDS3},
 
-	// VEX.128/256.66.0F38.WIG — sign/zero extend and broadcast (reg=dst, rm=src,
+	// VEX.128/256.66.0F38.WIG, sign/zero extend and broadcast (reg=dst, rm=src,
 	// no vvvv).
 	"VPMOVSXWD":    {2, 0x23, 0, 1, -1, vexRM},
 	"VPMOVSXDQ":    {2, 0x25, 0, 1, -1, vexRM},
@@ -143,70 +143,70 @@ var vexTable = map[string]vexSpec{
 	"VPBROADCASTQ": {2, 0x59, 0, 1, -1, vexRM},
 	"VPBROADCASTB": {2, 0x78, 0, 1, -1, vexRM},
 	"VPBROADCASTW": {2, 0x79, 0, 1, -1, vexRM},
-	// VEX.128/256.F3.0F.WIG — signed dword to packed double conversion
+	// VEX.128/256.F3.0F.WIG, signed dword to packed double conversion
 	// (reg=dst, rm=src, no vvvv; the length follows the destination).
 	"VCVTDQ2PD": {1, 0xE6, 0, 2, -1, vexRM},
-	// VEX.128/256.0F.WIG — signed dword to packed single conversion
+	// VEX.128/256.0F.WIG, signed dword to packed single conversion
 	// (reg=dst, rm=src, no vvvv, no mandatory prefix).
 	"VCVTDQ2PS": {1, 0x5B, 0, 0, -1, vexRM},
-	// VEX.128/256.0F.WIG — packed single to packed double conversion
+	// VEX.128/256.0F.WIG, packed single to packed double conversion
 	// (reg=dst, rm=src; the destination is the wide operand and sets the
 	// length).  Intel's maps prescribe the F3 prefix here (VEX.pp = 10), but
 	// the Go assembler emits the instruction with pp = 00, and gasm follows
-	// the Go assembler's bytes — its machine code is the oracle, not the
+	// the Go assembler's bytes, its machine code is the oracle, not the
 	// manual.
 	"VCVTPS2PD": {1, 0x5A, 0, 0, -1, vexRM},
-	// VEX.128.F2.0F.WIG — duplicate the low double of each 128-bit lane
+	// VEX.128.F2.0F.WIG, duplicate the low double of each 128-bit lane
 	// (reg=dst, rm=src, no vvvv; the length follows the destination).
 	"VMOVDDUP": {1, 0x12, 0, 3, -1, vexRM},
-	// VEX.128/256.66.0F.WIG — move mask to a GPR (reg=gpr dst, rm=vec src).
+	// VEX.128/256.66.0F.WIG, move mask to a GPR (reg=gpr dst, rm=vec src).
 	"VPMOVMSKB": {1, 0xD7, 0, 1, -1, vexRM},
 	"VMOVMSKPS": {1, 0x50, 0, 0, -1, vexRM}, // no 66 prefix (that would be VMOVMSKPD)
 
-	// VEX.128/256.66.0F.WIG — immediate shifts (opdigit selects the shift).
+	// VEX.128/256.66.0F.WIG, immediate shifts (opdigit selects the shift).
 	"VPSLLD": {1, 0x72, 0, 1, 6, vexShiftImm},
 	"VPSRAD": {1, 0x72, 0, 1, 4, vexShiftImm},
 	"VPSRLD": {1, 0x72, 0, 1, 2, vexShiftImm},
 	"VPSRLQ": {1, 0x73, 0, 1, 2, vexShiftImm},
 	"VPSLLQ": {1, 0x73, 0, 1, 6, vexShiftImm},
 
-	// VEX.128/256.66.0F.WIG — immediate shuffle (reg=dst, rm=src, imm8).
+	// VEX.128/256.66.0F.WIG, immediate shuffle (reg=dst, rm=src, imm8).
 	"VPSHUFD": {1, 0x70, 0, 1, -1, vexImmRM},
-	// VEX.256.66.0F3A.W1 — qword permute (reg=dst, rm=src, imm8).
+	// VEX.256.66.0F3A.W1, qword permute (reg=dst, rm=src, imm8).
 	"VPERMQ": {3, 0x00, 1, 1, -1, vexImmRM},
 
-	// VEX.128/256.66.0F.WIG — two-source shuffle (reg=dst, vvvv=src1, rm=src2,
+	// VEX.128/256.66.0F.WIG, two-source shuffle (reg=dst, vvvv=src1, rm=src2,
 	// imm8).
 	"VSHUFPD": {1, 0xC6, 0, 1, -1, vexNDS3Imm},
-	// VEX.256.66.0F3A.W0 — permute / insert (same shape; VINSERTI128's rm is
+	// VEX.256.66.0F3A.W0, permute / insert (same shape; VINSERTI128's rm is
 	// the XMM or memory source).
 	"VPERM2I128":  {3, 0x46, 0, 1, -1, vexNDS3Imm},
 	"VINSERTI128": {3, 0x38, 0, 1, -1, vexNDS3Imm},
 
-	// VEX.256.66.0F3A.W0 — lane extract (reg=YMM src, rm=XMM/memory dst, imm8).
+	// VEX.256.66.0F3A.W0, lane extract (reg=YMM src, rm=XMM/memory dst, imm8).
 	"VEXTRACTI128": {3, 0x39, 0, 1, -1, vexExtract},
 	"VEXTRACTF128": {3, 0x19, 0, 1, -1, vexExtract},
-	// VEX.128/256.66.0F3A.W0 — half-precision convert back ($imm, src, dst:
-	// reg=src, rm=XMM/memory dst, imm8 — the extract layout).
+	// VEX.128/256.66.0F3A.W0, half-precision convert back ($imm, src, dst:
+	// reg=src, rm=XMM/memory dst, imm8, the extract layout).
 	"VCVTPS2PH": {3, 0x1D, 0, 1, -1, vexExtract},
 
-	// VEX.128.0F.W0 — no operands.
+	// VEX.128.0F.W0, no operands.
 	"VZEROUPPER": {1, 0x77, 0, 0, -1, vexZero},
 
-	// VEX.128.0F.W0 — mask-register test (KTESTW k1, k2: reg = dst, rm = src).
+	// VEX.128.0F.W0, mask-register test (KTESTW k1, k2: reg = dst, rm = src).
 	"KTESTW": {1, 0x99, 0, 0, -1, vexRM},
 
-	// VEX.66.0F38.W0 — broadcast a single/double to all lanes (reg=dst,
+	// VEX.66.0F38.W0, broadcast a single/double to all lanes (reg=dst,
 	// rm=scalar memory; SD is 256-bit only).
 	"VBROADCASTSS": {2, 0x18, 0, 1, -1, vexRM},
 	"VBROADCASTSD": {2, 0x19, 0, 1, -1, vexRM},
-	// VEX.66.0F38.W0 — half-precision convert (reg=dst, rm=half-width
+	// VEX.66.0F38.W0, half-precision convert (reg=dst, rm=half-width
 	// source).
 	"VCVTPH2PS": {2, 0x13, 0, 1, -1, vexRM},
-	// VEX.F3.0F.WIG — replicate even/odd singles (reg=dst, rm=src).
+	// VEX.F3.0F.WIG, replicate even/odd singles (reg=dst, rm=src).
 	"VMOVSLDUP": {1, 0x12, 0, 2, -1, vexRM},
 	"VMOVSHDUP": {1, 0x16, 0, 2, -1, vexRM},
-	// VEX.66.0F.WIG — packed double to packed single conversion, the X/Y
+	// VEX.66.0F.WIG, packed double to packed single conversion, the X/Y
 	// spellings: the destination is always XMM and the spelling fixes the
 	// source length (X = 128, Y = 256).
 	"VCVTPD2PSX": {1, 0x5A, 0, 1, -1, vexRMSrcLen},
@@ -230,14 +230,14 @@ var vexTable = map[string]vexSpec{
 	"VCVTSI2SSL": {1, 0x2A, 0, 2, -1, vexNDS3},
 	"VCVTSI2SSQ": {1, 0x2A, 1, 2, -1, vexNDS3},
 
-	// VEX.128/256.66.0F.WIG — word shifts (opdigit selects the shift).
+	// VEX.128/256.66.0F.WIG, word shifts (opdigit selects the shift).
 	"VPSRLW": {1, 0x71, 0, 1, 2, vexShiftImm},
 	"VPSRAW": {1, 0x71, 0, 1, 4, vexShiftImm},
 	"VPSLLW": {1, 0x71, 0, 1, 6, vexShiftImm},
 
-	// VEX.F2.0F — packed double to packed dword conversions, truncating and
+	// VEX.F2.0F, packed double to packed dword conversions, truncating and
 	// non-truncating.  The destination is always XMM; the X/Y spellings fix
-	// the source length (XMM/YMM), and VEX.L follows it — see vexSrcLen.
+	// the source length (XMM/YMM), and VEX.L follows it, see vexSrcLen.
 	"VCVTPD2DQX":  {1, 0xE6, 0, 3, -1, vexRMSrcLen},
 	"VCVTPD2DQY":  {1, 0xE6, 0, 3, -1, vexRMSrcLen},
 	"VCVTTPD2DQX": {1, 0xE6, 0, 1, -1, vexRMSrcLen},
@@ -257,7 +257,7 @@ var vexSrcLen = map[string]int{
 	"VCVTPD2PSY":  1,
 }
 
-// vexVarShift maps the shift mnemonics to their variable-count opcode — the
+// vexVarShift maps the shift mnemonics to their variable-count opcode, the
 // form whose count comes from an XMM register or memory (VPSRLQ X0, Y8, Y8),
 // an ordinary NDS encoding rather than the /digit immediate form above.
 var vexVarShift = map[string]byte{
@@ -288,20 +288,20 @@ type vexMoveSpec struct {
 
 // vexMoveTable maps an upper-case move mnemonic to its encoding.
 var vexMoveTable = map[string]vexMoveSpec{
-	// VEX.128/256.F3.0F.WIG — unaligned integer move.
+	// VEX.128/256.F3.0F.WIG, unaligned integer move.
 	"VMOVDQU": {1, 2, 0x6F, 0x7F, 0, 0, 0, 0, true, false, false},
-	// VEX.128/256.66.0F.WIG — unaligned packed double move.
+	// VEX.128/256.66.0F.WIG, unaligned packed double move.
 	"VMOVUPD": {1, 1, 0x10, 0x11, 0, 0, 0, 0, true, false, false},
-	// VEX.128.66.0F.W0 — 32-bit GPR/memory ↔ XMM.
+	// VEX.128.66.0F.W0, 32-bit GPR/memory ↔ XMM.
 	"VMOVD": {1, 1, 0x6E, 0x7E, 0, 0, 0, 0, false, true, true},
-	// VMOVQ — 66 6E W1 (r/m→xmm), 66 7E W1 (xmm→r/m), 66 D6 W0 (xmm→xmm).
+	// VMOVQ, 66 6E W1 (r/m→xmm), 66 7E W1 (xmm→r/m), 66 D6 W0 (xmm→xmm).
 	"VMOVQ": {1, 1, 0x6E, 0x7E, 1, 1, 0xD6, 0, true, true, true},
-	// VEX.128.F2.0F.WIG — scalar double move, memory operands only (the
+	// VEX.128.F2.0F.WIG, scalar double move, memory operands only (the
 	// register form takes three operands and is not supported yet).
 	"VMOVSD": {1, 3, 0x10, 0x11, 0, 0, 0, 0, false, false, true},
-	// VEX.128.F3.0F.WIG — scalar single move, memory operands only.
+	// VEX.128.F3.0F.WIG, scalar single move, memory operands only.
 	"VMOVSS": {1, 2, 0x10, 0x11, 0, 0, 0, 0, false, false, true},
-	// VEX.128/256 — aligned packed moves.
+	// VEX.128/256, aligned packed moves.
 	"VMOVAPS": {1, 0, 0x28, 0x29, 0, 0, 0, 0, true, false, false},
 	"VMOVAPD": {1, 1, 0x28, 0x29, 0, 0, 0, 0, true, false, false},
 }
@@ -317,7 +317,7 @@ func isVex(mnemUpper string) bool {
 
 // encodeVex encodes a VEX instruction with operands in Plan 9 order.
 func (e *enc) encodeVex(mnemUpper string, ops []Operand) error {
-	// Vector register indices 16–31 exist only in EVEX encodings; fail
+	// Vector register indices 16-31 exist only in EVEX encodings; fail
 	// loudly rather than silently truncating the index.
 	for _, op := range ops {
 		if r, ok := op.(Reg); ok && r.isVec() && r.idx >= 16 {
@@ -420,7 +420,7 @@ func (e *enc) encodeVexRM(spec vexSpec, ops []Operand) error {
 }
 
 // encodeVexRMSrcLen encodes a length-narrowing conversion: OP src, dst with
-// the destination always XMM and the VEX.L bit following the source — fixed
+// the destination always XMM and the VEX.L bit following the source, fixed
 // by the mnemonic's spelling (VCVTPD2DQX = 128, VCVTPD2DQY = 256) even when
 // the source is memory.
 func (e *enc) encodeVexRMSrcLen(mnem string, spec vexSpec, ops []Operand) error {
diff --git a/ast/ast.go b/ast/ast.go
index 346ad3e..e0f5718 100644
--- a/ast/ast.go
+++ b/ast/ast.go
@@ -17,7 +17,7 @@ type File struct {
 	Orphans []Stmt // labels/instructions seen before any TEXT directive
 	// Macros holds the names introduced by #define directives in this file.
 	// The linter uses it to avoid flagging macro invocations as unknown
-	// instructions (macro expansion itself is out of scope — see the docs).
+	// instructions (macro expansion itself is out of scope, see the docs).
 	Macros map[string]bool
 }
 
diff --git a/format/format.go b/format/format.go
index 1fcae06..fbc9ac0 100644
--- a/format/format.go
+++ b/format/format.go
@@ -1,7 +1,7 @@
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
 // SPDX-License-Identifier: BSD-3-Clause
 
-// Package format implements a canonical formatter for GAsm source — the
+// Package format implements a canonical formatter for GAsm source, the
 // equivalent of gofmt for Plan 9 assembly.  It works on the token stream
 // rather than the AST so that every line (including comments and blanks) is
 // preserved; it only normalises indentation, operand spacing and per-function
@@ -92,7 +92,7 @@ func Source(src string) string {
 		case kInstr:
 			out = renderInstr(line, maxWidth[inf.funcID])
 			// A RET ends the body for indentation purposes: comments that
-			// follow it — typically the next function's doc comment — belong
+			// follow it, typically the next function's doc comment, belong
 			// at column 0, not inside the finished function.
 			if strings.EqualFold(line[0].Text, "RET") {
 				inBody = false
@@ -120,8 +120,8 @@ type outLine struct {
 }
 
 // normalizeSpacing enforces the canonical blank-line layout: runs of blank
-// lines collapse to one, and a new block — a label, or a TEXT or GLOBL
-// directive — is preceded by exactly one blank line.  Comments immediately
+// lines collapse to one, and a new block, a label, or a TEXT or GLOBL
+// directive, is preceded by exactly one blank line.  Comments immediately
 // above a block belong to it, so the blank line is inserted before them.  No
 // blank line is forced at the top of the file, right after a TEXT (the
 // function's first label), or between stacked labels that share an address.
diff --git a/verify/abi_amd64.go b/verify/abi_amd64.go
index cdd107a..bd4a432 100644
--- a/verify/abi_amd64.go
+++ b/verify/abi_amd64.go
@@ -38,7 +38,7 @@ func enterJITChecked(fn uintptr, stack uintptr)
 
 // leaveJITCheckedRaw is the raw return trampoline for ABI checks.  Its
 // address is obtained from the GLOBL in abi_amd64.s (leaveCheckedPtr),
-// which points to the .abi0 code — NOT the ABIInternal wrapper that this
+// which points to the .abi0 code, NOT the ABIInternal wrapper that this
 // declaration would generate.  The declaration exists solely to satisfy
 // go vet's "missing Go declaration" check.
 //
diff --git a/verify/abi_arm64.go b/verify/abi_arm64.go
index 1e4576c..db079b5 100644
--- a/verify/abi_arm64.go
+++ b/verify/abi_arm64.go
@@ -22,7 +22,7 @@ func enterJITChecked(fn uintptr, stack uintptr)
 
 // leaveJITCheckedRaw is the raw return trampoline for ABI checks.  Its
 // address is obtained from the GLOBL in abi_arm64.s (leaveCheckedPtr),
-// which points to the .abi0 code — NOT the ABIInternal wrapper that this
+// which points to the .abi0 code, NOT the ABIInternal wrapper that this
 // declaration would generate.  The declaration exists solely to satisfy
 // go vet's "missing Go declaration" check.
 //
diff --git a/verify/abi_loong64.go b/verify/abi_loong64.go
index 7f9474e..efafcf0 100644
--- a/verify/abi_loong64.go
+++ b/verify/abi_loong64.go
@@ -22,7 +22,7 @@ func enterJITChecked(fn uintptr, stack uintptr)
 
 // leaveJITCheckedRaw is the raw return trampoline for ABI checks.  Its
 // address is obtained from the GLOBL in abi_loong64.s (leaveCheckedPtr),
-// which points to the .abi0 code — NOT the ABIInternal wrapper that this
+// which points to the .abi0 code, NOT the ABIInternal wrapper that this
 // declaration would generate.  The declaration exists solely to satisfy
 // go vet's "missing Go declaration" check.
 //
diff --git a/verify/abi_riscv64.go b/verify/abi_riscv64.go
index 6ea9ed5..1be06b4 100644
--- a/verify/abi_riscv64.go
+++ b/verify/abi_riscv64.go
@@ -22,7 +22,7 @@ func enterJITChecked(fn uintptr, stack uintptr)
 
 // leaveJITCheckedRaw is the raw return trampoline for ABI checks.  Its
 // address is obtained from the GLOBL in abi_riscv64.s (leaveCheckedPtr),
-// which points to the .abi0 code — NOT the ABIInternal wrapper that this
+// which points to the .abi0 code, NOT the ABIInternal wrapper that this
 // declaration would generate.  The declaration exists solely to satisfy
 // go vet's "missing Go declaration" check.
 //
diff --git a/verify/call_amd64.go b/verify/call_amd64.go
index ae05dcf..8acd2c5 100644
--- a/verify/call_amd64.go
+++ b/verify/call_amd64.go
@@ -57,7 +57,7 @@ const stackPad = 64
 // (the ABI0 convention shares the argument area for inputs and outputs).
 //
 // The function must be NOSPLIT (no stack growth) and must not reference
-// external symbols — the image is self-contained.
+// external symbols, the image is self-contained.
 func Call(fnAddr uintptr, args []byte) ([]byte, error) {
 	// Prepare the stack: [padding][leaveJIT addr][args...]
 	stackSize := stackPad + 8 + len(args) + 64 // padding + ret + args + safety
diff --git a/verify/fuzz.go b/verify/fuzz.go
index 986e839..1985060 100644
--- a/verify/fuzz.go
+++ b/verify/fuzz.go
@@ -33,7 +33,7 @@ func (r FuzzResult) String() string {
 	if r.OK() {
 		return fmt.Sprintf("%s: %d/%d iterations match", r.Func, r.Matches, r.Iterations)
 	}
-	s := fmt.Sprintf("%s: %d/%d match, %d MISMATCH — %s",
+	s := fmt.Sprintf("%s: %d/%d match, %d MISMATCH: %s",
 		r.Func, r.Matches, r.Iterations, r.Mismatches, r.FirstFail)
 	if len(r.CrashInput) > 0 {
 		s += fmt.Sprintf("\n  input: %x", r.CrashInput)
@@ -295,7 +295,7 @@ func genDualArgs(rng *rand.Rand, sig funcSig, argSize int) (gasmArgs, goArgs []b
 			bufs = append(bufs, buf1, buf2)
 			putPtr(gasmArgs, off, unsafe.Pointer(&buf1[0]))
 			putPtr(goArgs, off, unsafe.Pointer(&buf2[0]))
-			// len and cap both equal declaredLen — the buffer is guaranteed
+			// len and cap both equal declaredLen, the buffer is guaranteed
 			// to hold at least declaredLen elements plus safety margin.
 			putU64(gasmArgs, off+8, uint64(declaredLen))
 			putU64(gasmArgs, off+16, uint64(declaredLen))
diff --git a/verify/fuzz_extra_test.go b/verify/fuzz_extra_test.go
index 29cbd0e..526d246 100644
--- a/verify/fuzz_extra_test.go
+++ b/verify/fuzz_extra_test.go
@@ -19,7 +19,7 @@ func TestFuzzResultString(t *testing.T) {
 	t.Run("mismatch", func(t *testing.T) {
 		r := FuzzResult{Func: "mul", Iterations: 100, Matches: 95, Mismatches: 5, FirstFail: "iter 23"}
 		s := r.String()
-		if s != "mul: 95/100 match, 5 MISMATCH — iter 23" {
+		if s != "mul: 95/100 match, 5 MISMATCH: iter 23" {
 			t.Errorf("String() = %q", s)
 		}
 	})
@@ -27,7 +27,7 @@ func TestFuzzResultString(t *testing.T) {
 	t.Run("crash", func(t *testing.T) {
 		r := FuzzResult{Func: "dec", Iterations: 100, Matches: 99, Mismatches: 1, FirstFail: "SIGSEGV", CrashInput: []byte{0x01, 0x02}}
 		s := r.String()
-		if s != "dec: 99/100 match, 1 MISMATCH — SIGSEGV\n  input: 0102" {
+		if s != "dec: 99/100 match, 1 MISMATCH: SIGSEGV\n  input: 0102" {
 			t.Errorf("String() = %q", s)
 		}
 	})