feat(cmd): GOOS-aware headers, audit battery shapes and semicolon spacing

Assisted-by: GLM 5.3 Flash

asm/assemble.go

@@ -846,6 +846,14 @@ func operandFromAST(op *ast.Operand, size int, fi frameInfo, link *linkInfo) (Op
 	case ast.OpAddr:
 		a := op.Addr
 
+		// A bracketed register range, [Z0-Z3]: the four-register source of
+		// the 4FMAPS/4VNNIW families.  The EVEX quad-register emit path
+		// needs an encoder operand of its own, so the shape stays a named
+		// gap rather than an encoding.
+		if a.Range != nil {
+			return nil, fmt.Errorf("register range %q needs quad-register encoder support", op.Raw)
+		}
+
 		// FP-relative: x+N(FP) → (N + fpAdjust)(SP).  The offset N lives in the
 		// symbol, not the address displacement.
 		if a.Sym != nil && a.Sym.Pseudo == "FP" {
@@ -893,6 +901,16 @@ func operandFromAST(op *ast.Operand, size int, fi frameInfo, link *linkInfo) (Op
 			}
 			return m, nil
 		}
+		// Index-only memory: the VSIB form the gather/scatter families
+		// read, 8(X4*1).  A scaled vector index addresses memory with no
+		// base register; the mod=00 SIB with base field 101 carries it.
+		if a.Index != "" {
+			idx, ok := ParseReg(a.Index)
+			if !ok {
+				return nil, fmt.Errorf("unknown index register %q", a.Index)
+			}
+			return Mem{Index: idx, Scale: a.Scale, Disp: a.Offset, HasIndex: true, Size: size}, nil
+		}
 		// Bare register.
 		if a.Sym != nil && a.Sym.Pseudo == "" && a.Sym.Name != "" {
 			if r, ok := ParseReg(a.Sym.Name); ok {

ast/ast.go

@@ -151,11 +151,21 @@ type Immediate struct {
 // Address is a non-immediate operand: a register, a memory reference, a symbol
 // reference or a label.  Fields are populated best-effort from the syntax.
 type Address struct {
-	Sym    *Symbol // name reference (bare ident, or name+off(pseudo))
+	Sym    *Symbol   // name reference (bare ident, or name+off(pseudo))
-	Base   string  // base register, from (base)
+	Base   string    // base register, from (base)
-	Index  string  // index register, from (index*scale)
+	Index  string    // index register, from (index*scale)
-	Scale  int     // index scale; 0 when absent
+	Scale  int       // index scale; 0 when absent
-	Offset int64   // leading displacement, from off(base)
+	Offset int64     // leading displacement, from off(base)
-	HasOff bool    // a leading displacement is present
+	HasOff bool      // a leading displacement is present
-	Shift  string  // verbatim arm64 shift suffix, e.g. "<< 2"
+	Shift  string    // verbatim arm64 shift suffix, e.g. "<< 2"
+	Range  *RegRange // bracketed register range; nil for every other form
+}
+
+// RegRange is a bracketed register range, [Z0-Z3]: the amd64 spelling of
+// the four-register source of the 4FMAPS/4VNNIW families.  Lo and Hi carry
+// the verbatim register spellings; the range is inclusive at both ends.
+type RegRange struct {
+	Lo  string
+	Hi  string
+	Pos token.Position
 }

cmd/gasm/asmhdr.go

@@ -27,7 +27,10 @@ import (
 // GOROOT assembly includes it, and a standalone assembler has no compiler
 // to have produced it, so gasm generates the equivalent itself: the package
 // the .s file lives in is parsed and type-checked here, with the target
-// architecture's own sizes, and the same defines are written out.
+// architecture's own sizes, and the same defines are written out.  The
+// type-checking GOOS is selected by the caller: a GOOS-specific file
+// (sys_darwin_arm64.s) needs its platform's defines, which a header from
+// the ambient GOOS silently omits.
 //
 // The emitter mirrors cmd/compile's dumpasmhdr exactly: constants come out
 // as "const_NAME", struct entries as "NAME__size" followed by the fields in
@@ -67,11 +70,15 @@ func goAsmHeaderResolved(asmDir string, dirs []string) bool {
 	return false
 }
 
-// generateGoAsmHeader type-checks the Go package in pkgDir for goarch,
+// generateGoAsmHeader type-checks the Go package in pkgDir for goos and
-// writes its go_asm.h equivalent into dir, and returns dir.  The caller
+// goarch, writes its go_asm.h equivalent into dir, and returns dir.  An
-// owns the directory and its removal.
+// empty goos means the ambient one.  The caller owns the directory and its
-func generateGoAsmHeader(pkgDir, goarch, dir string) (string, error) {
+// removal.
-	imp := newSourceImporter(goarch)
+func generateGoAsmHeader(pkgDir, goos, goarch, dir string) (string, error) {
+	if goos == "" {
+		goos = build.Default.GOOS
+	}
+	imp := newSourceImporter(goos, goarch)
 	if imp.sizes == nil {
 		return "", fmt.Errorf("go_asm.h: unknown GOARCH %q", goarch)
 	}
@@ -89,7 +96,7 @@ func generateGoAsmHeader(pkgDir, goarch, dir string) (string, error) {
 	}
 
 	var b strings.Builder
-	fmt.Fprintf(&b, "// generated by gasm from package %s (GOARCH %s)\n\n", bp.Name, goarch)
+	fmt.Fprintf(&b, "// generated by gasm from package %s (GOOS %s, GOARCH %s)\n\n", bp.Name, goos, goarch)
 	// Files in the build's own order and declarations in source order: the
 	// same walk the compiler's reader makes, so the header reads the same
 	// way the toolchain's does.  Order carries no meaning to the assembler
@@ -203,13 +210,14 @@ type sourceImporter struct {
 	pkgs  map[string]*types.Package
 }
 
-// newSourceImporter returns the importer for one target architecture.
+// newSourceImporter returns the importer for one target GOOS and GOARCH.
 // Cgo is disabled so the file set is deterministic and independent of the
 // host's C toolchain: cgo-tagged files drop out of the build exactly as
 // they do from a CGO_ENABLED=0 build, whose assembly is what gasm targets.
-func newSourceImporter(goarch string) *sourceImporter {
+func newSourceImporter(goos, goarch string) *sourceImporter {
 	ctxt := new(build.Context)
 	*ctxt = build.Default
+	ctxt.GOOS = goos
 	ctxt.GOARCH = goarch
 	ctxt.CgoEnabled = false
 	return &sourceImporter{
@@ -291,7 +299,7 @@ func (im *sourceImporter) checkPackage(bp *build.Package, files []*ast.File) (*t
 // type-check must not be re-checked once per file.
 type asmhdrCache struct {
 	root string
-	dirs map[string]string // "pkgDir\x00goarch" -> directory holding go_asm.h
+	dirs map[string]string // "pkgDir\x00goos\x00goarch" -> directory holding go_asm.h
 	errs map[string]error
 }
 
@@ -304,17 +312,22 @@ func newAsmhdrCache() (*asmhdrCache, error) {
 }
 
 // dirFor returns the directory holding the generated go_asm.h for pkgDir
-// and goarch, generating it on first use.
+// under goos and goarch, generating it on first use.  An empty goos means
-func (c *asmhdrCache) dirFor(pkgDir, goarch string) (string, error) {
+// the ambient one, resolved here so that one package cannot generate twice
-	key := pkgDir + "\x00" + goarch
+// under an explicit and an implicit spelling of the same GOOS.
+func (c *asmhdrCache) dirFor(pkgDir, goos, goarch string) (string, error) {
+	if goos == "" {
+		goos = build.Default.GOOS
+	}
+	key := pkgDir + "\x00" + goos + "\x00" + goarch
 	if dir, ok := c.dirs[key]; ok {
 		return dir, nil
 	}
 	if err, ok := c.errs[key]; ok {
 		return "", err
 	}
-	dir := filepath.Join(c.root, fmt.Sprintf("h%d_%s", len(c.dirs), goarch))
+	dir := filepath.Join(c.root, fmt.Sprintf("h%d_%s_%s", len(c.dirs), goos, goarch))
-	if _, err := generateGoAsmHeader(pkgDir, goarch, dir); err != nil {
+	if _, err := generateGoAsmHeader(pkgDir, goos, goarch, dir); err != nil {
 		c.errs[key] = err
 		return "", err
 	}
@@ -327,10 +340,10 @@ func (c *asmhdrCache) close() { os.RemoveAll(c.root) }
 
 // ensureGoAsmHeader prepares the include directory a file that includes
 // go_asm.h needs: the generated header for the package in path's directory,
-// for the file's target architecture.  It reports a usage error when the
+// for the file's target GOOS and architecture.  It reports a usage error
-// architecture cannot be determined, and passes through the generator's
+// when the architecture cannot be determined, and passes through the
-// diagnostics, which name the package.
+// generator's diagnostics, which name the package.
-func ensureGoAsmHeader(path string, target arch.Arch, cache *asmhdrCache) (string, func(), error) {
+func ensureGoAsmHeader(path string, target arch.Arch, goos string, cache *asmhdrCache) (string, func(), error) {
 	if path == "-" {
 		return "", nil, errors.New("cannot generate go_asm.h for standard input (no package directory)")
 	}
@@ -338,14 +351,14 @@ func ensureGoAsmHeader(path string, target arch.Arch, cache *asmhdrCache) (strin
 		return "", nil, errors.New("a file that includes go_asm.h needs a target architecture: name the file _<arch>.s or pass -GOARCH")
 	}
 	if cache != nil {
-		dir, err := cache.dirFor(filepath.Dir(path), goarchName(target))
+		dir, err := cache.dirFor(filepath.Dir(path), goos, goarchName(target))
 		return dir, func() {}, err
 	}
 	root, err := os.MkdirTemp("", "gasm-asmhdr")
 	if err != nil {
 		return "", nil, err
 	}
-	dir, err := generateGoAsmHeader(filepath.Dir(path), goarchName(target), root)
+	dir, err := generateGoAsmHeader(filepath.Dir(path), goos, goarchName(target), root)
 	if err != nil {
 		os.RemoveAll(root)
 		return "", nil, err

cmd/gasm/audit.go

@@ -5,6 +5,7 @@ package main
 
 import (
 	"fmt"
+	"maps"
 	"os"
 	"os/exec"
 	"path/filepath"
@@ -288,6 +289,11 @@ func probeShapes(a arch.Arch) []string {
 			"V1.B16, [V2.B16], V3.B16", "V1.B8, [V2.B16, V3.B16], V4.B8",
 			"$4, V1.B16, V2.B16, V3.B16", "$15, V1", "V1, V2, p2",
 			"R0, R1, $1, $4, p2",
+			// The landing-pad kind, the compiler's PCDATA
+			// bookkeeping and the four-operand bitfield
+			// insert/extract family, as the toolchain's own
+			// testdata spells them.
+			"C", "$1, $0", "$0, R1, $1, R2",
 		}
 	case arch.RISCV:
 		return []string{
@@ -307,6 +313,9 @@ func probeShapes(a arch.Arch) []string {
 			"X5, X6, p2", "R5, R6, p2",
 			"X5, E8, M8, TA, MA, X6", "$4, E32, M1, TA, MA, X1",
 			"(X5), X6, V1, V2",
+			// The CSR immediate forms the toolchain's testdata spells:
+			// immediate, CSR name, destination.
+			"$2, TIME, X5",
 			"",
 		}
 	case arch.LOONG64:
@@ -323,6 +332,12 @@ func probeShapes(a arch.Arch) []string {
 			"V1, V2, V3", "X1, X2, X3", "V1, V2", "X1, X2", "V1", "X1",
 			// The vector compare-to-flag forms land in an FCC register.
 			"V1, FCC0", "X1, FCC0",
+			// The compiler's bookkeeping pair and the raw spellings the
+			// toolchain's own testdata carries: JIRL rd, rj, offset (the
+			// form RET lowers to), the prefetch with a 32-bit address and
+			// hint, and the byte-shuffle quads.
+			"$1, $0", "R1, R5, 0", "0(R7), $5, $0", "(R7), $5, $0",
+			"V1, V2, V3, V4", "X1, X2, X3, X4",
 			"",
 		}
 	}
@@ -487,10 +502,47 @@ func otherPortFile(path string) bool {
 
 // goOSNames are the GOOS values go/build recognises in file names.
 var goOSNames = map[string]bool{
-	"aix": true, "darwin": true, "dragonfly": true, "freebsd": true,
+	"aix": true, "android": true, "darwin": true, "dragonfly": true,
-	"ios": true, "js": true, "linux": true, "netbsd": true,
+	"freebsd": true, "hurd": true, "illumos": true, "ios": true,
+	"js": true, "linux": true, "nacl": true, "netbsd": true,
 	"openbsd": true, "plan9": true, "solaris": true, "wasip1": true,
-	"windows": true,
+	"windows": true, "zos": true,
+}
+
+// resolveGOOS validates a -GOOS flag value, mirroring the architecture
+// check's surface: a usage error naming what the tool accepts.
+func resolveGOOS(name string) (string, error) {
+	lower := strings.ToLower(name)
+	if goOSNames[lower] {
+		return lower, nil
+	}
+	return "", &usageError{fmt.Errorf("unknown GOOS %q: want one of %s", name, strings.Join(slices.Sorted(maps.Keys(goOSNames)), ", "))}
+}
+
+// goosFromFilename returns the GOOS the file's name carries, by go/build's
+// goodOSArchFile rule: the GOOS segment sits last, or last before the
+// architecture segment (sys_darwin_arm64.s, vlop_arm.s carries none).  An
+// empty result means the name names no GOOS and the ambient one applies.
+func goosFromFilename(path string) string {
+	base := path
+	if i := strings.LastIndexByte(base, '/'); i >= 0 {
+		base = base[i+1:]
+	}
+	base = strings.TrimSuffix(base, ".s")
+	// go/build ignores everything before the first underscore, so a GOOS
+	// segment is only ever looked for from there on.
+	i := strings.IndexByte(base, '_')
+	if i < 0 {
+		return ""
+	}
+	segs := strings.Split(base[i:], "_")
+	if n := len(segs); n >= 2 && goOSNames[segs[n-2]] && slices.Contains(goPortSuffixes, segs[n-1]) {
+		return segs[n-2]
+	}
+	if goOSNames[segs[len(segs)-1]] {
+		return segs[len(segs)-1]
+	}
+	return ""
 }
 
 // otherGOOSFile reports whether the file's name names a GOOS other than the
@@ -543,6 +595,12 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 			return nil, err
 		}
 
+		// The GOOS the header generation type-checks under follows the
+		// file's name when the name carries one; the ambient GOOS is the
+		// honest guess otherwise (a build tag naming another GOOS is
+		// invisible to a file-name rule).
+		goos := goosFromFilename(path)
+
 		var wanted []int // indexes into targets
 		if a := arch.FromFilename(path); a != arch.Unknown {
 			for i, tg := range targets {
@@ -567,11 +625,12 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 
 		// A file that includes go_asm.h parses against a per-target header:
 		// the defines differ per architecture (internal/cpu's layout, for
-		// one), so the parse cannot be shared the way a header-free file's
+		// one) and per GOOS (sys_darwin_arm64.s's trampoline constants,
-		// can.  A generation failure is a failure for every target, named
+		// for another), so the parse cannot be shared the way a
-		// for the package rather than a bare "include not found".  A header
+		// header-free file's can.  A generation failure is a failure for
-		// already resolvable in the package directory or the -I list is
+		// every target, named for the package rather than a bare "include
-		// left alone.
+		// not found".  A header already resolvable in the package
+		// directory or the -I list is left alone.
 		if len(wanted) > 0 && needsGoAsmHeader(src) && !goAsmHeaderResolved(filepath.Dir(path), dirs) {
 			if hdr == nil {
 				if hdr, err = newAsmhdrCache(); err != nil {
@@ -583,7 +642,7 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 			for _, i := range wanted {
 				tg, t := targets[i], tallies[i]
 				t.attempted++
-				hdrDir, err := hdr.dirFor(pkgDir, goarchName(tg.a))
+				hdrDir, err := hdr.dirFor(pkgDir, goos, goarchName(tg.a))
 				if err != nil {
 					ok = false
 					t.fail(path, corpusReason(err))

cmd/gasm/main.go

@@ -486,7 +486,7 @@ hover, document symbols, diagnostics and semantic-token highlighting.
 }
 
 func cmdAsm(args []string) int {
-	fs := newCommand("asm", "gasm asm [--format raw|elf|goobj] [-I dir] [-p pkg] [-GOARCH arch] [-o out] <file>", `
+	fs := newCommand("asm", "gasm asm [--format raw|elf|goobj] [-I dir] [-p pkg] [-GOARCH arch] [-GOOS os] [-o out] <file>", `
 	Assemble FILE without the Go toolchain: every TEXT function is encoded to
 	machine code and printed as a hex dump.  Supported architectures: amd64
 	(including VEX/AVX2 and EVEX/AVX-512), arm64 (AArch64 integer, FP,
@@ -507,18 +507,22 @@ and the format version from go version).
 A file that includes go_asm.h gets that header generated automatically from
 the package it lives in (the .go files beside it, type-checked for the
 target architecture, the toolchain's own defines), so GOROOT assembly
-assembles without a compiler.  A package that has no Go files for the
+assembles without a compiler.  -GOOS selects the type-checking GOOS for
-target or does not type-check is a hard error naming the package.
+that header: a GOOS-specific file (sys_darwin_arm64.s) needs its platform's
+defines, which a header from the ambient GOOS silently omits.  A package
+that has no Go files for the target or does not type-check is a hard error
+naming the package.
 `)
 	out := fs.String("o", "", "write the output to this file")
 	format := fs.String("format", "raw", "output format: raw (concatenated image), elf or goobj (Go object)")
 	pkg := fs.String("p", "", "package path for --format goobj (qualifies the exported symbols)")
 	archName := fs.String("GOARCH", "", "target architecture: amd64, arm64, riscv64 or loong64 (overrides the file-name suffix)")
+	goosName := fs.String("GOOS", "", "operating system for go_asm.h generation: a GOOS go/build recognises (default: the host's)")
 	var dirs includeDirs
 	fs.Var(&dirs, "I", "directory to search for #include files (may be repeated)")
 	fs.Parse(args)
 	if fs.NArg() != 1 {
-		fmt.Fprintln(os.Stderr, "usage: gasm asm [--format raw|elf|goobj] [-I dir] [-p pkg] [-GOARCH arch] [-o out] <file>")
+		fmt.Fprintln(os.Stderr, "usage: gasm asm [--format raw|elf|goobj] [-I dir] [-p pkg] [-GOARCH arch] [-GOOS os] [-o out] <file>")
 		return 2
 	}
 	// The format is validated before anything else, so a bogus value exits 2
@@ -539,6 +543,15 @@ target or does not type-check is a hard error naming the package.
 		}
 		targetArch = a
 	}
+	goos := ""
+	if *goosName != "" {
+		g, err := resolveGOOS(*goosName)
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "gasm asm: %v\n", err)
+			return 2
+		}
+		goos = g
+	}
 	src, err := readSource(path)
 	if err != nil {
 		fmt.Fprintln(os.Stderr, "gasm:", err)
@@ -553,7 +566,7 @@ target or does not type-check is a hard error naming the package.
 	// A go_asm.h that already resolves (placed by hand, or passed with -I)
 	// is left alone.
 	if needsGoAsmHeader(src) && !goAsmHeaderResolved(filepath.Dir(path), dirs) {
-		hdrDir, cleanup, err := ensureGoAsmHeader(path, targetArch, nil)
+		hdrDir, cleanup, err := ensureGoAsmHeader(path, targetArch, goos, nil)
 		if err != nil {
 			fmt.Fprintln(os.Stderr, "gasm asm:", err)
 			return 1

docs/CLI.md

@@ -141,7 +141,7 @@ gasm lint kernel_amd64.s
 ## asm
 
 ```text
-Usage: gasm asm [--format raw|elf|goobj] [-I dir] [-p pkg] [-GOARCH arch] [-o out] <file>
+Usage: gasm asm [--format raw|elf|goobj] [-I dir] [-p pkg] [-GOARCH arch] [-GOOS os] [-o out] <file>
 ```
 
 | Flag | Default | Effect |
@@ -150,6 +150,7 @@ Usage: gasm asm [--format raw|elf|goobj] [-I dir] [-p pkg] [-GOARCH arch] [-o ou
 | `-I` | empty | directory to search for `#include` files; may be repeated, searched in order after the source directory |
 | `-p` | empty | package path for `--format goobj`, qualifying the exported symbols |
 | `-GOARCH` | empty | target architecture: `amd64`, `arm64`, `riscv64` or `loong64`; overrides the file-name suffix |
+| `-GOOS` | empty | operating system for the generated `go_asm.h`: any GOOS `go/build` recognises in file names; default is the host's |
 | `-o` | empty | write the output to this file instead of a hex dump on stdout |
 
 Supported architectures: amd64 (VEX/AVX2 and EVEX/AVX-512 included), arm64,
@@ -163,6 +164,13 @@ system toolchain; `goobj` emits the Go toolchain's own object format, which
 installed: the object preamble is captured from `go tool asm` and the format
 version from `go version`. `raw` and `elf` need no toolchain at all.
 
+A file that includes `go_asm.h` gets that header generated from the Go
+files beside it, type-checked for the target. `-GOOS` selects the
+type-checking GOOS for that header, because a GOOS-specific file needs its
+platform's defines: `sys_darwin_arm64.s` fails against the ambient GOOS
+(`machTimebaseInfo_numer` is missing from a linux type-check) and assembles
+with `-GOOS darwin`.
+
 Assembly preprocessing matches the toolchain's: `#define` macros (object and
 parameterised) expand at the point of use, `#undef`, `#ifdef`, `#ifndef`,
 `#else` and `#endif` behave as in `go tool asm`, `;` separates statements,
@@ -390,7 +398,9 @@ With `--corpus` the audit changes shape: it assembles every `.s` file under
 DIR (default `GOROOT/src`) with the gasm encoder only, no toolchain probing.
 A file whose name carries a recognisable `_arch` suffix is attempted for that
 architecture; a file without one is attempted for all four, exactly as a
-`GOARCH` build would compile it. The report gives the headline number (files
+`GOARCH` build would compile it, and a name that names a GOOS
+(`sys_darwin_arm64.s`) type-checks its generated `go_asm.h` for that GOOS.
+The report gives the headline number (files
 that assemble for every target architecture), the per-architecture pass rates
 and the most common failure reasons with one representative file each, which
 drive the encodability backlog by frequency rather than by table order. A run

docs/man/gasm-asm.1

@@ -2,7 +2,7 @@
 .SH NAME
 gasm-asm \- assemble Plan 9 assembly without the Go toolchain
 .SH SYNOPSIS
-.B gasm asm [\-\-format raw|elf|goobj] [\-I dir] [\-p pkg] [\-GOARCH arch] [\-o out] <file>
+.B gasm asm [\-\-format raw|elf|goobj] [\-I dir] [\-p pkg] [\-GOARCH arch] [\-GOOS os] [\-o out] <file>
 .SH DESCRIPTION
 Assemble FILE without the Go toolchain: every TEXT function is encoded
 to machine code and printed as a hex dump.  Supported architectures:
@@ -42,6 +42,15 @@ need no toolchain at all.
 Framed functions receive the stack-split guard and the trailing
 morestack block, byte-identical to the toolchain's output, so split
 functions link too.
+.PP
+A file that includes go_asm.h gets that header generated from the Go
+files beside it, type-checked for the target.
+.B \-GOOS
+selects the type-checking GOOS for that header, because a GOOS-specific
+file needs its platform's defines: sys_darwin_arm64.s fails against the
+ambient GOOS (machTimebaseInfo_numer is missing from a linux type-check)
+and assembles with
+.BR "\-GOOS darwin" .
 .SH OPTIONS
 .TP
 .B \-\-format \fIraw|elf|goobj\fR
@@ -59,6 +68,10 @@ Target architecture: amd64, arm64, riscv64 or loong64; overrides the
 file-name suffix, which is how the suffix-less majority of GOROOT's
 files (cpu_x86.s, stub.s, ...) become assemblable.
 .TP
+.B \-GOOS \fIos\fR
+Operating system for the generated go_asm.h: any GOOS go/build
+recognises in file names; the default is the host's.
+.TP
 .B \-o \fIfile\fR
 Write the output to this file instead of a hex dump on stdout.
 .SH EXIT STATUS

format/format.go

@@ -241,6 +241,13 @@ func renderInstr(line []token.Token, width int) string {
 	if line[0].Kind != token.Ident {
 		return "\t" + mnem + " " + ops
 	}
+	// A statement separator belongs to the statement it ends: when the
+	// operands open with a ';', the alignment padding would land between
+	// the mnemonic and its own separator (REP  ; MOVSQ), so such a line
+	// renders with a single space whatever the function's width.
+	if strings.HasPrefix(ops, ";") {
+		return "\t" + mnem + " " + ops
+	}
 	if width < len(mnem) {
 		width = len(mnem)
 	}

format/format_test.go

@@ -364,6 +364,17 @@ func TestSemicolonSeparators(t *testing.T) {
 			in:   "TEXT ·f(SB), $0\nBYTE $1;\nRET\n",
 			want: "TEXT ·f(SB), $0\n\tBYTE $1;\n\tRET\n",
 		},
+		{
+			// The REP shape: a prefix-style zero-operand statement
+			// followed by the instruction it prefixes.  The separator
+			// belongs to the statement it ends, so the function's
+			// alignment width (MOVSQ is the widest mnemonic here) must
+			// not open a gap before it: one space after the mnemonic
+			// whatever the neighbours' lengths.
+			name: "after a prefix-style statement",
+			in:   "TEXT ·f(SB), $0\nMOVQ AX, BX\nREP; MOVSQ\nRET\n",
+			want: "TEXT ·f(SB), $0\n\tMOVQ AX, BX\n\tREP ; MOVSQ\n\tRET\n",
+		},
 	}
 	for _, tc := range cases {
 		t.Run(tc.name, func(t *testing.T) {

parser/parser.go

@@ -304,7 +304,7 @@ func (p *state) parseGlobl(line []token.Token) *ast.Globl {
 		rest = rest[1:]
 	}
 	if len(rest) > 0 && rest[0].Kind == token.Dollar {
-		g.Size = parseOperand(rest)
+		g.Size = parseOperand(rest, false)
 	}
 	return g
 }
@@ -321,7 +321,7 @@ func (p *state) parseData(line []token.Token) *ast.Data {
 	d.Name = sym
 	d.Width = width
 	if len(valuePart) > 0 {
-		d.Value = parseOperand(stripComment(valuePart))
+		d.Value = parseOperand(stripComment(valuePart), false)
 	}
 	return d
 }
@@ -332,8 +332,13 @@ func (p *state) parseInstr(line []token.Token) {
 		return
 	}
 	instr := &ast.Instr{Mnemonic: body[0], Comment: comment}
-	for _, grp := range splitOperands(body[1:]) {
+	grps := splitOperands(body[1:])
-		if op := parseOperand(grp); op != nil {
+	for i, grp := range grps {
+		// Only the final operand slot may carry a bare constant: the
+		// toolchain reads the trailing 1 of CMPSD X1, X0, 1 as $1
+		// (math/floor_amd64.s), while an earlier bare number names an
+		// absolute address, a form this parser keeps out of the tree.
+		if op := parseOperand(grp, i == len(grps)-1); op != nil {
 			instr.Operands = append(instr.Operands, op)
 		}
 	}
@@ -417,8 +422,10 @@ func setName(raw string, sym *ast.Symbol) {
 
 // --- operand parsing --------------------------------------------------------
 
-// parseOperand parses one operand group into an Operand.
+// parseOperand parses one operand group into an Operand.  allowBare marks
-func parseOperand(g []token.Token) *ast.Operand {
+// the final operand slot of an instruction, where the toolchain reads a
+// bare constant expression as an immediate.
+func parseOperand(g []token.Token, allowBare bool) *ast.Operand {
 	g = stripComment(g)
 	if len(g) == 0 {
 		return nil
@@ -431,9 +438,25 @@ func parseOperand(g []token.Token) *ast.Operand {
 	}
 	op.Kind = ast.OpAddr
 	op.Addr = parseAddress(g)
+	// A trailing bare constant leaves every address field empty: the
+	// grammar sees no register, memory reference or symbol, and the closed
+	// constant expression is the whole group.  Read it as the immediate it
+	// names, exactly what the $ spelling would produce.
+	if allowBare && isEmptyAddress(op.Addr) {
+		if v, rest, ok := foldExpr(g); ok && len(rest) == 0 {
+			op.Kind = ast.OpImmediate
+			op.Imm = ast.Immediate{Val: v, HasVal: true}
+		}
+	}
 	return op
 }
 
+// isEmptyAddress reports whether parseAddress populated nothing, its sign
+// that the group is no register, memory reference, symbol or register range.
+func isEmptyAddress(a ast.Address) bool {
+	return a.Sym == nil && a.Base == "" && a.Index == "" && a.Range == nil && a.Shift == ""
+}
+
 // parseImmediate parses the tokens following a '$'.
 func parseImmediate(g []token.Token) ast.Immediate {
 	var imm ast.Immediate
@@ -497,6 +520,14 @@ func parseAddress(g []token.Token) ast.Address {
 	if len(g) == 0 {
 		return addr
 	}
+	// A bracketed register range, [Z0-Z3]: the amd64 4FMAPS/4VNNIW
+	// multi-source operand.  The bracket runes arrive as Illegal tokens
+	// (the lexer has no bracket kind), so the shape matches on their text.
+	if isBracket(g[0], "[") && len(g) == 5 && g[1].Kind == token.Ident &&
+		g[2].Kind == token.Minus && g[3].Kind == token.Ident && isBracket(g[4], "]") {
+		addr.Range = &ast.RegRange{Lo: g[1].Text, Hi: g[3].Text, Pos: g[0].Pos}
+		return addr
+	}
 	// Symbol-with-pseudo form: name[<>][+off](PSEUDO).
 	// When the prefix is not a valid symbol name (e.g. a bare number like
 	// 0(SP) in RISC-V), sym is nil, and we fall through to regular memory
@@ -577,6 +608,16 @@ func parseAddress(g []token.Token) ast.Address {
 			}
 		}
 	}
+	// A lone (index*scale) group is the VSIB index-only form: the
+	// gather/scatter families address memory through a scaled vector index
+	// with no base register, 8(X4*1).  The two-group grammar below reads
+	// (base)(index*scale), so a first group whose member carries a scale
+	// factor can only be an index.
+	if isIndexGroup(g[i:]) {
+		addr.Index = g[i+1].Text
+		addr.Scale = int(parseInt(g[i+3].Text))
+		i += 5
+	}
 	// First parenthesised group: the base register.
 	if i < len(g) && g[i].Kind == token.LParen {
 		i++
@@ -633,6 +674,19 @@ func findPseudoParen(g []token.Token) int {
 	return -1
 }
 
+// isBracket reports whether t is a square bracket.  The lexer has no bracket
+// kind, so '[' and ']' arrive as Illegal tokens.
+func isBracket(t token.Token, text string) bool {
+	return t.Kind == token.Illegal && t.Text == text
+}
+
+// isIndexGroup reports whether g begins with a complete (index*scale) group:
+// one identifier followed by a scale factor, all inside a single parenthesis.
+func isIndexGroup(g []token.Token) bool {
+	return len(g) >= 5 && g[0].Kind == token.LParen && g[1].Kind == token.Ident &&
+		g[2].Kind == token.Star && g[3].Kind == token.Number && g[4].Kind == token.RParen
+}
+
 // --- token helpers ----------------------------------------------------------
 
 // splitOperands splits a token slice on top-level commas (commas outside any

parser/parser_test.go

@@ -497,3 +497,140 @@ func TestParseEqualsZeroOptions(t *testing.T) {
 		}
 	}
 }
+
+// TestBracketRegisterRange pins the amd64 multi-source operand of the
+// 4FMAPS/4VNNIW families: the bracket group [Z0-Z3] names four consecutive
+// source registers and must reach the AST as a register range instead of an
+// empty address.
+func TestBracketRegisterRange(t *testing.T) {
+	file, errs := Parse("t.s", "TEXT \u00b7f(SB), NOSPLIT, $0\n\tV4FMADDPS 17(SP), [Z0-Z3], K2, Z0\n\tRET\n")
+	if len(errs) > 0 {
+		t.Fatalf("parse errors: %v", errs)
+	}
+	fn := file.Decls[0].(*ast.Text)
+	in := fn.Body[0].(*ast.Instr)
+	if len(in.Operands) != 4 {
+		t.Fatalf("operands = %d, want 4", len(in.Operands))
+	}
+	rng := in.Operands[1]
+	if rng.Kind != ast.OpAddr {
+		t.Errorf("range operand kind = %v, want OpAddr", rng.Kind)
+	}
+	if rng.Addr.Range == nil {
+		t.Fatalf("range operand = %+v, want a register range", rng.Addr)
+	}
+	if rng.Addr.Range.Lo != "Z0" || rng.Addr.Range.Hi != "Z3" {
+		t.Errorf("range = %s-%s, want Z0-Z3", rng.Addr.Range.Lo, rng.Addr.Range.Hi)
+	}
+	if rng.Addr.Sym != nil || rng.Addr.Base != "" || rng.Addr.Index != "" || rng.Addr.Shift != "" {
+		t.Errorf("range operand carries stray address fields: %+v", rng.Addr)
+	}
+	if rng.Raw != "[ Z0 - Z3 ]" {
+		t.Errorf("range raw = %q, want the verbatim spelling", rng.Raw)
+	}
+}
+
+// TestBracketRegisterRangeNotList pins that arm64-style register lists, whose
+// members carry arrangements, stay out of the simple range shape: they remain
+// plain bracketed groups the arm64 encoder reads from Raw.  A comma inside
+// brackets is a top-level comma, so a multi-member list spans several
+// operands, exactly the shape the arm64 encoder's list scan stitches back.
+func TestBracketRegisterRangeNotList(t *testing.T) {
+	file, errs := Parse("t.s", "TEXT \u00b7f(SB), NOSPLIT, $0\n\tVLD1 (R2), [V21.B16]\n\tVLD1 (R1), [V2.B16, V3.B16]\n\tRET\n")
+	if len(errs) > 0 {
+		t.Fatalf("parse errors: %v", errs)
+	}
+	fn := file.Decls[0].(*ast.Text)
+	for i, want := range []string{"[ V21.B16 ]", "V3.B16 ]"} {
+		in := fn.Body[i].(*ast.Instr)
+		op := in.Operands[len(in.Operands)-1]
+		if op.Addr.Range != nil {
+			t.Errorf("%s: range = %v, want nil", in.Mnemonic.Text, op.Addr.Range)
+		}
+		if op.Raw != want {
+			t.Errorf("operand %d raw = %q, want %q", i, op.Raw, want)
+		}
+	}
+}
+
+// TestVSIBIndexOnly pins the gather/scatter memory operand with a scaled
+// vector index and no base register: 8(X4*1) must carry index and scale and
+// leave the base empty, not strand the scale in the shift suffix.
+func TestVSIBIndexOnly(t *testing.T) {
+	file, errs := Parse("t.s", "TEXT \u00b7f(SB), NOSPLIT, $0\n\tVPGATHERDQ Y0, 8(X4*1), Y6\n\tVPGATHERDQ Y0, (X4*2), Y6\n\tVPGATHERDQ Y0, -8(X4*1), Y6\n\tRET\n")
+	if len(errs) > 0 {
+		t.Fatalf("parse errors: %v", errs)
+	}
+	fn := file.Decls[0].(*ast.Text)
+	want := []ast.Address{
+		{Index: "X4", Scale: 1, Offset: 8, HasOff: true},
+		{Index: "X4", Scale: 2},
+		{Index: "X4", Scale: 1, Offset: -8, HasOff: true},
+	}
+	for i, w := range want {
+		in := fn.Body[i].(*ast.Instr)
+		a := in.Operands[1].Addr
+		if a.Base != "" || a.Index != w.Index || a.Scale != w.Scale || a.Offset != w.Offset || a.HasOff != w.HasOff || a.Shift != "" {
+			t.Errorf("operand %d = %+v, want %+v", i, a, w)
+		}
+	}
+}
+
+// TestVSIBTwoGroupKeepsBase pins that the ordinary (base)(index*scale)
+// grammar is untouched by the index-only recognition.
+func TestVSIBTwoGroupKeepsBase(t *testing.T) {
+	file, errs := Parse("t.s", "TEXT \u00b7f(SB), NOSPLIT, $0\n\tVP4DPWSSD 7(SI)(DI*1), [Z2-Z5], K4, Z17\n\tRET\n")
+	if len(errs) > 0 {
+		t.Fatalf("parse errors: %v", errs)
+	}
+	fn := file.Decls[0].(*ast.Text)
+	in := fn.Body[0].(*ast.Instr)
+	a := in.Operands[0].Addr
+	if a.Base != "SI" || a.Index != "DI" || a.Scale != 1 || a.Offset != 7 || !a.HasOff {
+		t.Errorf("address = %+v, want base SI index DI scale 1 offset 7", a)
+	}
+	if in.Operands[1].Addr.Range == nil || in.Operands[1].Addr.Range.Lo != "Z2" || in.Operands[1].Addr.Range.Hi != "Z5" {
+		t.Errorf("second operand = %+v, want range Z2-Z5", in.Operands[1].Addr)
+	}
+}
+
+// TestBareTrailingImmediate pins the toolchain's bare constant spelling in
+// the final operand slot: CMPSD X1, X0, 1 reads as $1 (math/floor_amd64.s).
+// Earlier slots keep the strict grammar, so a bare number there stays an
+// address rather than becoming an immediate.
+func TestBareTrailingImmediate(t *testing.T) {
+	file, errs := Parse("t.s", "TEXT \u00b7f(SB), NOSPLIT, $0\n\tCMPSD X1, X0, 1\n\tCMPSD X1, X0, -1\n\tADDQ AX, 1+2\n\tRET\n")
+	if len(errs) > 0 {
+		t.Fatalf("parse errors: %v", errs)
+	}
+	fn := file.Decls[0].(*ast.Text)
+	for i, want := range []int64{1, -1, 3} {
+		in := fn.Body[i].(*ast.Instr)
+		last := in.Operands[len(in.Operands)-1]
+		if last.Kind != ast.OpImmediate || !last.Imm.HasVal || last.Imm.Val != want {
+			t.Errorf("operand %d = %+v, want immediate %d", i, last, want)
+		}
+	}
+
+	// A bare number outside the final slot is not an immediate.
+	file2, errs2 := Parse("t.s", "TEXT \u00b7f(SB), NOSPLIT, $0\n\tADDQ 1, AX\n\tRET\n")
+	if len(errs2) > 0 {
+		t.Fatalf("parse errors: %v", errs2)
+	}
+	fn2 := file2.Decls[0].(*ast.Text)
+	first := fn2.Body[0].(*ast.Instr).Operands[0]
+	if first.Kind != ast.OpAddr {
+		t.Errorf("non-final bare number kind = %v, want OpAddr", first.Kind)
+	}
+	// A bare name in the final slot stays a symbol: labels are names, not
+	// constants, and jump targets depend on the distinction.
+	file3, errs3 := Parse("t.s", "TEXT \u00b7f(SB), NOSPLIT, $0\n\tJMP loop\nloop: NOP\n\tRET\n")
+	if len(errs3) > 0 {
+		t.Fatalf("parse errors: %v", errs3)
+	}
+	fn3 := file3.Decls[0].(*ast.Text)
+	jmp := fn3.Body[0].(*ast.Instr)
+	if jmp.Operands[0].Kind != ast.OpAddr || jmp.Operands[0].Addr.Sym == nil || jmp.Operands[0].Addr.Sym.Name != "loop" {
+		t.Errorf("jump target = %+v, want label loop", jmp.Operands[0])
+	}
+}

parser/preproc.go

@@ -349,7 +349,7 @@ func (pp *preproc) expandTokens(in []token.Token) []token.Token {
 			consecutive = 0
 			continue
 		}
-		def := pp.macros[t.Text]
+		def, suffix := pp.macroFor(t.Text)
 		if def == nil {
 			i++
 			consecutive = 0
@@ -363,7 +363,14 @@ func (pp *preproc) expandTokens(in []token.Token) []token.Token {
 			return nil
 		}
 		if def.args == nil {
-			s = append(s[:i], append(restamp(def.body, t.Pos), s[i+1:]...)...)
+			body := restamp(def.body, t.Pos)
+			if suffix != "" {
+				// The macro was reached only through a compound spelling
+				// (ACC0.B16 over "#define ACC0 V8"), so the selector has
+				// to travel with the expansion.
+				body = appendSelector(body, suffix, t.Pos)
+			}
+			s = append(s[:i], append(body, s[i+1:]...)...)
 			continue
 		}
 		// A parameterised macro invoked without its parentheses stands
@@ -394,6 +401,17 @@ func (pp *preproc) expandTokens(in []token.Token) []token.Token {
 					sub = append(sub, restamp(args[k], t.Pos)...)
 					continue
 				}
+				// A parameter used with an element or lane selector: the
+				// lexer folds A.S4 into one identifier, so the whole-token
+				// match above cannot see the parameter.  The toolchain
+				// lexes the period separately and substitutes the name
+				// alone; splitting at the FIRST period and pasting the
+				// argument back in front of the selector is the equivalent
+				// for this lexer.
+				if k, sel := parameterSelector(bt.Text, def.args); k >= 0 {
+					sub = append(sub, restamp(pasteSelector(args[k], sel), t.Pos)...)
+					continue
+				}
 			}
 			sub = append(sub, bt)
 		}
@@ -402,6 +420,75 @@ func (pp *preproc) expandTokens(in []token.Token) []token.Token {
 	return s
 }
 
+// macroFor finds the macro a use names.  The lexer folds NAME.selector into
+// one identifier token, so a macro written behind a selector suffix
+// (ACC0.B16 over "#define ACC0 V8") never matches a whole-token table
+// lookup; the toolchain splits on the period and reads the two halves, so
+// the prefix before the FIRST period is tried here as well and the caller
+// re-attaches the suffix to whatever the macro expands to.  Only a whole
+// name counts: AB.S4 does not reach a macro named A, and a parameterised
+// macro is not hidden behind a selector, because its invocation would need
+// the parentheses to follow the bare name.
+func (pp *preproc) macroFor(text string) (*macroDef, string) {
+	if def := pp.macros[text]; def != nil {
+		return def, ""
+	}
+	if j := strings.IndexByte(text, '.'); j > 0 {
+		if def := pp.macros[text[:j]]; def != nil && def.args == nil {
+			return def, text[j:]
+		}
+	}
+	return nil, ""
+}
+
+// appendSelector glues a selector suffix onto an object macro's expansion:
+// the selector binds to the identifier the expansion ends with, the way the
+// toolchain's operand parser reads V0 and .B16 back as one register
+// spelling.  An expansion that does not end in an identifier carries the
+// selector as its own token, which the parser then reports where it cannot
+// parse it.
+func appendSelector(body []token.Token, suffix string, pos token.Position) []token.Token {
+	if n := len(body); n > 0 && body[n-1].Kind == token.Ident {
+		body[n-1].Text += suffix
+		return body
+	}
+	return append(body, token.Token{Kind: token.Ident, Text: suffix, Pos: pos, End: pos})
+}
+
+// parameterSelector reports the argument a compound body token names: the
+// parameter whose whole name occupies the text before the token's FIRST
+// period, with the selector that follows.  k is negative when no parameter
+// matches, which leaves tokens like AB.S4 untouched even though a parameter
+// A is bound.
+func parameterSelector(text string, args []string) (int, string) {
+	j := strings.IndexByte(text, '.')
+	if j <= 0 {
+		return -1, ""
+	}
+	if k := slices.Index(args, text[:j]); k >= 0 {
+		return k, text[j:]
+	}
+	return -1, ""
+}
+
+// pasteSelector joins an argument with the selector a compound body token
+// carries, textually: the selector binds to the identifier the argument
+// ends with, so A.S4 over the argument V0.B16 spells V0.B16.S4, exactly the
+// operand the toolchain's split-then-substitute leaves behind.  An argument
+// with no trailing identifier carries the selector as a separate token,
+// which the parser then reports where it cannot parse it.
+func pasteSelector(val []token.Token, suffix string) []token.Token {
+	if len(val) == 0 {
+		return []token.Token{{Kind: token.Ident, Text: suffix}}
+	}
+	out := slices.Clone(val)
+	if n := len(out); out[n-1].Kind == token.Ident {
+		out[n-1].Text += suffix
+		return out
+	}
+	return append(out, token.Token{Kind: token.Ident, Text: suffix})
+}
+
 // collectArgs reads the actual argument tokens of an invocation; the opening
 // parenthesis is at start.  Commas separate arguments except inside nested
 // parentheses.  A nil result means the list was unterminated, which is a

parser/preproc_test.go

@@ -536,6 +536,127 @@ func TestConstantExpressionFoldsWithoutExpand(t *testing.T) {
 	}
 }
 
+func TestParameterWithSelectorSubstitutes(t *testing.T) {
+	// The lexer folds A.S4 into one identifier token, so a parameter used
+	// with an element or lane selector never matched the whole-token
+	// substitution; the toolchain's lexer splits on the period and its
+	// substitution sees the name alone.  Several parameters carry selectors
+	// in one body here, which is the chacha8_arm64.s QR shape in miniature.
+	_, got := expand(t, `
+#define QR(A, B, C, D) VADD A.S4, B.S4, C.S4; VEOR D.B16, A.B16, D.B16
+TEXT ·f(SB), NOSPLIT, $0
+	QR(V0, V1, V2, V3)
+	RET
+`)
+	wantLines(t, got,
+		"VADD V0.S4, V1.S4, V2.S4",
+		"VEOR V3.B16, V0.B16, V3.B16",
+		"RET",
+	)
+}
+
+func TestSelectorWithCompoundArgumentPastesTextually(t *testing.T) {
+	// An argument that is itself one compound identifier pastes verbatim:
+	// A.S4 over V0.B16 spells V0.B16.S4, the operand the toolchain's
+	// split-then-substitute leaves behind.
+	_, got := expand(t, `
+#define M(A) VADD A.S4, A.S4, A.S4
+TEXT ·f(SB), NOSPLIT, $0
+	M(V0.B16)
+	RET
+`)
+	wantLines(t, got, "VADD V0.B16.S4, V0.B16.S4, V0.B16.S4", "RET")
+}
+
+func TestSelectorAlongsideBareParameter(t *testing.T) {
+	// A body may use the parameter bare and suffixed, and the argument may
+	// itself end in a selector; neither disturbs the other.
+	_, got := expand(t, `
+#define M(A) VADD A, A.S4, A
+TEXT ·f(SB), NOSPLIT, $0
+	M(V0)
+	M(V1.B16)
+	RET
+`)
+	wantLines(t, got,
+		"VADD V0, V0.S4, V0",
+		"VADD V1.B16, V1.B16.S4, V1.B16",
+		"RET",
+	)
+}
+
+func TestSelectorKeepsNonParameterPrefixes(t *testing.T) {
+	// The prefix before the period must be the whole parameter name:
+	// AB.S4 never reaches a parameter A.
+	_, got := expand(t, `
+#define M(A) VADD AB.S4, A.S4, AB.S4
+TEXT ·f(SB), NOSPLIT, $0
+	M(V0)
+	RET
+`)
+	wantLines(t, got, "VADD AB.S4, V0.S4, AB.S4", "RET")
+}
+
+func TestSelectorExpandsMacroValuedArgument(t *testing.T) {
+	// gcm_arm64.s invokes mulRound(B1) where B1 is itself an object macro:
+	// the paste stays rescannable, so B1.D1 still expands to V1.D1 the way
+	// the toolchain's rescan of substituted tokens does.
+	_, got := expand(t, `
+#define B1 V1
+#define mulRound(X) VPMULL X.D1, T1.D1, T3.Q1
+TEXT ·f(SB), NOSPLIT, $0
+	mulRound(B1)
+	RET
+`)
+	wantLines(t, got, "VPMULL V1.D1, T1.D1, T3.Q1", "RET")
+}
+
+func TestObjectMacroBehindSelectorExpands(t *testing.T) {
+	// Ordinary code writes ACC0.B16 where ACC0 is an object macro; the
+	// toolchain expands the alias because its lexer reads the selector as
+	// its own token, and the lookup here must reach the macro through the
+	// compound spelling the same way.
+	_, got := expand(t, `
+#define ACC0 V8
+TEXT ·f(SB), NOSPLIT, $0
+	VEOR ACC0.B16, ACC0.B16, ACC0.B16
+	RET
+`)
+	wantLines(t, got, "VEOR V8.B16, V8.B16, V8.B16", "RET")
+}
+
+func TestChacha8QRMacroExpands(t *testing.T) {
+	// The real QR round of chacha8_arm64.s end to end: every parameter
+	// carries a selector somewhere, and the round is sixteen instructions.
+	_, got := expand(t, `
+#define QR(A, B, C, D) \
+	VADD A.S4, B.S4, A.S4; VEOR D.B16, A.B16, D.B16; VREV32 D.H8, D.H8; \
+	VADD C.S4, D.S4, C.S4; VEOR B.B16, C.B16, V30.B16; VSHL $12, V30.S4, B.S4; VSRI $20, V30.S4, B.S4; \
+	VADD A.S4, B.S4, A.S4; VEOR D.B16, A.B16, D.B16; VTBL V31.B16, [D.B16], D.B16; \
+	VADD C.S4, D.S4, C.S4; VEOR B.B16, C.B16, V30.B16; VSHL $7, V30.S4, B.S4; VSRI $25, V30.S4, B.S4
+TEXT ·f(SB), NOSPLIT, $0
+	QR(V0, V1, V2, V3)
+	RET
+`)
+	wantLines(t, got,
+		"VADD V0.S4, V1.S4, V0.S4",
+		"VEOR V3.B16, V0.B16, V3.B16",
+		"VREV32 V3.H8, V3.H8",
+		"VADD V2.S4, V3.S4, V2.S4",
+		"VEOR V1.B16, V2.B16, V30.B16",
+		"VSHL $12, V30.S4, V1.S4",
+		"VSRI $20, V30.S4, V1.S4",
+		"VADD V0.S4, V1.S4, V0.S4",
+		"VEOR V3.B16, V0.B16, V3.B16",
+		"VTBL V31.B16, [V3.B16], V3.B16",
+		"VADD V2.S4, V3.S4, V2.S4",
+		"VEOR V1.B16, V2.B16, V30.B16",
+		"VSHL $7, V30.S4, V1.S4",
+		"VSRI $25, V30.S4, V1.S4",
+		"RET",
+	)
+}
+
 func TestNotAnExpressionFallsBack(t *testing.T) {
 	// Symbol immediates and floats must keep their ordinary parse.
 	f, errs := ParseWithOptions("t_amd64.s", "TEXT ·f(SB), NOSPLIT, $0\n\tMOVQ $1.5, AX\n\tMOVQ $·sym(SB), AX\n\tRET\n", Options{})

testdata/verify/vsib_amd64.s

@@ -0,0 +1,21 @@
+#include "textflag.h"
+
+// The index-only VSIB shapes from amd64enc_extra.s: a scaled vector index
+// addresses memory with no base register.  The comments record the bytes the
+// Go assembler emits.
+TEXT asmtest(SB), DUPOK|NOSPLIT, $0
+	VPGATHERDQ Y0, 8(X4*1), Y6 // c4e2fd90342508000000
+	VPGATHERDQ Y0,-8(X4*1), Y6 // c4e2fd903425f8ffffff
+	VPGATHERDQ Y0, 0(X4*1), Y6 // c4e2fd90342500000000
+	VPGATHERDQ Y0, 664(X4*1), Y6 // c4e2fd90342598020000
+	VPGATHERDQ Y0, 8(X14*1), Y6 // c4a2fd90343508000000
+	VPGATHERDQ Y0, 8(X4*8), Y6 // c4e2fd9034e508000000
+	VGATHERDPD Y0, 8(X4*1), Y6 // c4e2fd92342508000000
+	VPGATHERDD X0, 8(X4*1), X6 // c4e27d90342508000000
+	VPGATHERDD Y0, 8(Y4*1), Y6 // c4e2fd90342508000000
+
+	// The bare trailing immediate of math/floor_amd64.s: the toolchain
+	// reads the final 1 as $1.
+	CMPSD      X1, X0, 1 // f20fc2c101
+	CMPSD      X1, X0, 2 // f20fc2c102
+	RET

verify/groundtruth_test.go

@@ -123,9 +123,11 @@ func TestGroundTruthAMD64(t *testing.T) {
 		"../testdata/verify/sse_amd64.s",
 		"../testdata/verify/avx_amd64.s",
 		"../testdata/verify/pfx_amd64.s",
+		"../testdata/verify/vsib_amd64.s",
 		"../testdata/verify/rawdata_amd64.s",
 		"../testdata/verify/avx512_amd64.s",
 		"../testdata/verify/pfx_amd64.s",
+		"../testdata/verify/vsib_amd64.s",
 		"../testdata/verify/rawdata_amd64.s",
 		"../testdata/verify/avx512_amd64.s",
 		"../testdata/verify/doubleshift_amd64.s",