Exemple #1
0
func main() {
	log.SetFlags(0)
	log.SetPrefix("asm: ")

	GOARCH := obj.Getgoarch()

	architecture := arch.Set(GOARCH)
	if architecture == nil {
		log.Fatalf("asm: unrecognized architecture %s", GOARCH)
	}

	flags.Parse()

	// Create object file, write header.
	fd, err := os.Create(*flags.OutputFile)
	if err != nil {
		log.Fatal(err)
	}
	ctxt := obj.Linknew(architecture.LinkArch)
	if *flags.PrintOut {
		ctxt.Debugasm = 1
	}
	ctxt.LineHist.TrimPathPrefix = *flags.TrimPath
	ctxt.Flag_dynlink = *flags.Dynlink
	if *flags.Shared || *flags.Dynlink {
		ctxt.Flag_shared = 1
	}
	ctxt.Bso = obj.Binitw(os.Stdout)
	defer ctxt.Bso.Flush()
	output := obj.Binitw(fd)
	fmt.Fprintf(output, "go object %s %s %s\n", obj.Getgoos(), obj.Getgoarch(), obj.Getgoversion())
	fmt.Fprintf(output, "!\n")

	lexer := lex.NewLexer(flag.Arg(0), ctxt)
	parser := asm.NewParser(ctxt, architecture, lexer)
	diag := false
	ctxt.DiagFunc = func(format string, args ...interface{}) {
		diag = true
		log.Printf(format, args...)
	}
	pList := obj.Linknewplist(ctxt)
	var ok bool
	pList.Firstpc, ok = parser.Parse()
	if ok {
		// reports errors to parser.Errorf
		obj.Writeobjdirect(ctxt, output)
	}
	if !ok || diag {
		log.Printf("asm: assembly of %s failed", flag.Arg(0))
		os.Remove(*flags.OutputFile)
		os.Exit(1)
	}
	output.Flush()
}
Exemple #2
0
		}

		// zero old range
		p = zerorange(p, int64(frame), lo, hi)

		// set new range
		hi = n.Xoffset + n.Type.Width

		lo = n.Xoffset
	}

	// zero final range
	zerorange(p, int64(frame), lo, hi)
}

var darwin = obj.Getgoos() == "darwin"

func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
	cnt := hi - lo
	if cnt == 0 {
		return p
	}
	if cnt < int64(4*gc.Widthptr) {
		for i := int64(0); i < cnt; i += int64(gc.Widthptr) {
			p = appendpp(p, arm64.AMOVD, obj.TYPE_REG, arm64.REGZERO, 0, obj.TYPE_MEM, arm64.REGSP, 8+frame+lo+i)
		}
	} else if cnt <= int64(128*gc.Widthptr) && !darwin { // darwin ld64 cannot handle BR26 reloc with non-zero addend
		p = appendpp(p, arm64.AMOVD, obj.TYPE_REG, arm64.REGSP, 0, obj.TYPE_REG, arm64.REGRT1, 0)
		p = appendpp(p, arm64.AADD, obj.TYPE_CONST, 0, 8+frame+lo-8, obj.TYPE_REG, arm64.REGRT1, 0)
		p.Reg = arm64.REGRT1
		p = appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_MEM, 0, 0)
Exemple #3
0
func Main() {
	if obj.Getgoos() == "nacl" {
		resvd = append(resvd, x86.REG_BP, x86.REG_R15)
	} else if obj.Framepointer_enabled != 0 {
		resvd = append(resvd, x86.REG_BP)
	}

	gc.Thearch.Thechar = thechar
	gc.Thearch.Thestring = thestring
	gc.Thearch.Thelinkarch = thelinkarch
	gc.Thearch.Typedefs = typedefs
	gc.Thearch.REGSP = x86.REGSP
	gc.Thearch.REGCTXT = x86.REGCTXT
	gc.Thearch.REGCALLX = x86.REG_BX
	gc.Thearch.REGCALLX2 = x86.REG_AX
	gc.Thearch.REGRETURN = x86.REG_AX
	gc.Thearch.REGMIN = x86.REG_AX
	gc.Thearch.REGMAX = x86.REG_R15
	gc.Thearch.FREGMIN = x86.REG_X0
	gc.Thearch.FREGMAX = x86.REG_X15
	gc.Thearch.MAXWIDTH = MAXWIDTH
	gc.Thearch.ReservedRegs = resvd

	gc.Thearch.AddIndex = addindex
	gc.Thearch.Betypeinit = betypeinit
	gc.Thearch.Cgen_bmul = cgen_bmul
	gc.Thearch.Cgen_hmul = cgen_hmul
	gc.Thearch.Cgen_shift = cgen_shift
	gc.Thearch.Clearfat = clearfat
	gc.Thearch.Defframe = defframe
	gc.Thearch.Dodiv = dodiv
	gc.Thearch.Excise = excise
	gc.Thearch.Expandchecks = expandchecks
	gc.Thearch.Getg = getg
	gc.Thearch.Gins = gins
	gc.Thearch.Ginsboolval = ginsboolval
	gc.Thearch.Ginscmp = ginscmp
	gc.Thearch.Ginscon = ginscon
	gc.Thearch.Ginsnop = ginsnop
	gc.Thearch.Gmove = gmove
	gc.Thearch.Linkarchinit = linkarchinit
	gc.Thearch.Peep = peep
	gc.Thearch.Proginfo = proginfo
	gc.Thearch.Regtyp = regtyp
	gc.Thearch.Sameaddr = sameaddr
	gc.Thearch.Smallindir = smallindir
	gc.Thearch.Stackaddr = stackaddr
	gc.Thearch.Blockcopy = blockcopy
	gc.Thearch.Sudoaddable = sudoaddable
	gc.Thearch.Sudoclean = sudoclean
	gc.Thearch.Excludedregs = excludedregs
	gc.Thearch.RtoB = RtoB
	gc.Thearch.FtoB = FtoB
	gc.Thearch.BtoR = BtoR
	gc.Thearch.BtoF = BtoF
	gc.Thearch.Optoas = optoas
	gc.Thearch.Doregbits = doregbits
	gc.Thearch.Regnames = regnames

	gc.Main()
	gc.Exit(0)
}
Exemple #4
0
func linknew(arch *LinkArch) *Link {
	ctxt := new(Link)
	ctxt.Hash = make(map[symVer]*LSym)
	ctxt.Arch = arch
	ctxt.Version = obj.HistVersion
	ctxt.Goroot = obj.Getgoroot()

	p := obj.Getgoarch()
	if p != arch.Name {
		log.Fatalf("invalid goarch %s (want %s)", p, arch.Name)
	}

	var buf string
	buf, _ = os.Getwd()
	if buf == "" {
		buf = "/???"
	}
	buf = filepath.ToSlash(buf)

	ctxt.Headtype = headtype(obj.Getgoos())
	if ctxt.Headtype < 0 {
		log.Fatalf("unknown goos %s", obj.Getgoos())
	}

	// Record thread-local storage offset.
	// TODO(rsc): Move tlsoffset back into the linker.
	switch ctxt.Headtype {
	default:
		log.Fatalf("unknown thread-local storage offset for %s", Headstr(ctxt.Headtype))

	case obj.Hplan9, obj.Hwindows:
		break

		/*
		 * ELF uses TLS offset negative from FS.
		 * Translate 0(FS) and 8(FS) into -16(FS) and -8(FS).
		 * Known to low-level assembly in package runtime and runtime/cgo.
		 */
	case obj.Hlinux,
		obj.Hfreebsd,
		obj.Hnetbsd,
		obj.Hopenbsd,
		obj.Hdragonfly,
		obj.Hsolaris:
		if obj.Getgoos() == "android" {
			switch ctxt.Arch.Thechar {
			case '6':
				// Android/amd64 constant - offset from 0(FS) to our TLS slot.
				// Explained in src/runtime/cgo/gcc_android_*.c
				ctxt.Tlsoffset = 0x1d0
			case '8':
				// Android/386 constant - offset from 0(GS) to our TLS slot.
				ctxt.Tlsoffset = 0xf8
			default:
				ctxt.Tlsoffset = -1 * ctxt.Arch.Ptrsize
			}
		} else {
			ctxt.Tlsoffset = -1 * ctxt.Arch.Ptrsize
		}

	case obj.Hnacl:
		switch ctxt.Arch.Thechar {
		default:
			log.Fatalf("unknown thread-local storage offset for nacl/%s", ctxt.Arch.Name)

		case '5':
			ctxt.Tlsoffset = 0

		case '6':
			ctxt.Tlsoffset = 0

		case '8':
			ctxt.Tlsoffset = -8
		}

		/*
		 * OS X system constants - offset from 0(GS) to our TLS.
		 * Explained in src/runtime/cgo/gcc_darwin_*.c.
		 */
	case obj.Hdarwin:
		switch ctxt.Arch.Thechar {
		default:
			log.Fatalf("unknown thread-local storage offset for darwin/%s", ctxt.Arch.Name)

		case '5':
			ctxt.Tlsoffset = 0 // dummy value, not needed

		case '6':
			ctxt.Tlsoffset = 0x8a0

		case '7':
			ctxt.Tlsoffset = 0 // dummy value, not needed

		case '8':
			ctxt.Tlsoffset = 0x468
		}
	}

	// On arm, record goarm.
	if ctxt.Arch.Thechar == '5' {
		ctxt.Goarm = obj.Getgoarm()
	}

	return ctxt
}
Exemple #5
0
func dumpobj() {
	var err error
	bout, err = obj.Bopenw(outfile)
	if err != nil {
		Flusherrors()
		fmt.Printf("can't create %s: %v\n", outfile, err)
		errorexit()
	}

	startobj := int64(0)
	var arhdr [ArhdrSize]byte
	if writearchive != 0 {
		obj.Bwritestring(bout, "!<arch>\n")
		arhdr = [ArhdrSize]byte{}
		bout.Write(arhdr[:])
		startobj = obj.Boffset(bout)
	}

	fmt.Fprintf(bout, "go object %s %s %s %s\n", obj.Getgoos(), obj.Getgoarch(), obj.Getgoversion(), obj.Expstring())
	dumpexport()

	if writearchive != 0 {
		bout.Flush()
		size := obj.Boffset(bout) - startobj
		if size&1 != 0 {
			obj.Bputc(bout, 0)
		}
		obj.Bseek(bout, startobj-ArhdrSize, 0)
		formathdr(arhdr[:], "__.PKGDEF", size)
		bout.Write(arhdr[:])
		bout.Flush()

		obj.Bseek(bout, startobj+size+(size&1), 0)
		arhdr = [ArhdrSize]byte{}
		bout.Write(arhdr[:])
		startobj = obj.Boffset(bout)
		fmt.Fprintf(bout, "go object %s %s %s %s\n", obj.Getgoos(), obj.Getgoarch(), obj.Getgoversion(), obj.Expstring())
	}

	if pragcgobuf != "" {
		if writearchive != 0 {
			// write empty export section; must be before cgo section
			fmt.Fprintf(bout, "\n$$\n\n$$\n\n")
		}

		fmt.Fprintf(bout, "\n$$  // cgo\n")
		fmt.Fprintf(bout, "%s\n$$\n\n", pragcgobuf)
	}

	fmt.Fprintf(bout, "\n!\n")

	externs := len(externdcl)

	dumpglobls()
	dumptypestructs()

	// Dump extra globals.
	tmp := externdcl

	if externdcl != nil {
		externdcl = externdcl[externs:]
	}
	dumpglobls()
	externdcl = tmp

	dumpdata()
	obj.Writeobjdirect(Ctxt, bout)

	if writearchive != 0 {
		bout.Flush()
		size := obj.Boffset(bout) - startobj
		if size&1 != 0 {
			obj.Bputc(bout, 0)
		}
		obj.Bseek(bout, startobj-ArhdrSize, 0)
		formathdr(arhdr[:], "_go_.o", size)
		bout.Write(arhdr[:])
	}

	obj.Bterm(bout)
}
Exemple #6
0
func Ldmain() {
	Ctxt = linknew(Thelinkarch)
	Ctxt.Thechar = int32(Thearch.Thechar)
	Ctxt.Thestring = Thestring
	Ctxt.Diag = Diag
	Ctxt.Bso = &Bso

	Bso = *obj.Binitw(os.Stdout)
	Debug = [128]int{}
	nerrors = 0
	outfile = ""
	HEADTYPE = -1
	INITTEXT = -1
	INITDAT = -1
	INITRND = -1
	INITENTRY = ""
	Linkmode = LinkAuto

	// For testing behavior of go command when tools crash silently.
	// Undocumented, not in standard flag parser to avoid
	// exposing in usage message.
	for _, arg := range os.Args {
		if arg == "-crash_for_testing" {
			os.Exit(2)
		}
	}

	if Thearch.Thechar == '6' && obj.Getgoos() == "plan9" {
		obj.Flagcount("8", "use 64-bit addresses in symbol table", &Debug['8'])
	}
	obj.Flagfn1("B", "add an ELF NT_GNU_BUILD_ID `note` when using ELF", addbuildinfo)
	obj.Flagcount("C", "check Go calls to C code", &Debug['C'])
	obj.Flagint64("D", "set data segment `address`", &INITDAT)
	obj.Flagstr("E", "set `entry` symbol name", &INITENTRY)
	obj.Flagfn1("I", "use `linker` as ELF dynamic linker", setinterp)
	obj.Flagfn1("L", "add specified `directory` to library path", Lflag)
	obj.Flagfn1("H", "set header `type`", setheadtype)
	obj.Flagint32("R", "set address rounding `quantum`", &INITRND)
	obj.Flagint64("T", "set text segment `address`", &INITTEXT)
	obj.Flagfn0("V", "print version and exit", doversion)
	obj.Flagfn1("X", "add string value `definition` of the form importpath.name=value", addstrdata1)
	obj.Flagcount("a", "disassemble output", &Debug['a'])
	obj.Flagstr("buildid", "record `id` as Go toolchain build id", &buildid)
	flag.Var(&Buildmode, "buildmode", "set build `mode`")
	obj.Flagcount("c", "dump call graph", &Debug['c'])
	obj.Flagcount("d", "disable dynamic executable", &Debug['d'])
	obj.Flagstr("extar", "archive program for buildmode=c-archive", &extar)
	obj.Flagstr("extld", "use `linker` when linking in external mode", &extld)
	obj.Flagstr("extldflags", "pass `flags` to external linker", &extldflags)
	obj.Flagcount("f", "ignore version mismatch", &Debug['f'])
	obj.Flagcount("g", "disable go package data checks", &Debug['g'])
	obj.Flagcount("h", "halt on error", &Debug['h'])
	obj.Flagstr("installsuffix", "set package directory `suffix`", &flag_installsuffix)
	obj.Flagstr("k", "set field tracking `symbol`", &tracksym)
	obj.Flagstr("libgcc", "compiler support lib for internal linking; use \"none\" to disable", &libgccfile)
	obj.Flagfn1("linkmode", "set link `mode` (internal, external, auto)", setlinkmode)
	flag.BoolVar(&Linkshared, "linkshared", false, "link against installed Go shared libraries")
	obj.Flagcount("msan", "enable MSan interface", &flag_msan)
	obj.Flagcount("n", "dump symbol table", &Debug['n'])
	obj.Flagstr("o", "write output to `file`", &outfile)
	flag.Var(&rpath, "r", "set the ELF dynamic linker search `path` to dir1:dir2:...")
	obj.Flagcount("race", "enable race detector", &flag_race)
	obj.Flagcount("s", "disable symbol table", &Debug['s'])
	var flagShared int
	if Thearch.Thechar == '5' || Thearch.Thechar == '6' {
		obj.Flagcount("shared", "generate shared object (implies -linkmode external)", &flagShared)
	}
	obj.Flagstr("tmpdir", "use `directory` for temporary files", &tmpdir)
	obj.Flagcount("u", "reject unsafe packages", &Debug['u'])
	obj.Flagcount("v", "print link trace", &Debug['v'])
	obj.Flagcount("w", "disable DWARF generation", &Debug['w'])

	obj.Flagstr("cpuprofile", "write cpu profile to `file`", &cpuprofile)
	obj.Flagstr("memprofile", "write memory profile to `file`", &memprofile)
	obj.Flagint64("memprofilerate", "set runtime.MemProfileRate to `rate`", &memprofilerate)

	// Clumsy hack to preserve old two-argument -X name val syntax for old scripts.
	// Rewrite that syntax into new syntax -X name=val.
	// TODO(rsc): Delete this hack in Go 1.6 or later.
	var args []string
	for i := 0; i < len(os.Args); i++ {
		arg := os.Args[i]
		if (arg == "-X" || arg == "--X") && i+2 < len(os.Args) && !strings.Contains(os.Args[i+1], "=") {
			fmt.Fprintf(os.Stderr, "link: warning: option %s %s %s may not work in future releases; use %s %s=%s\n",
				arg, os.Args[i+1], os.Args[i+2],
				arg, os.Args[i+1], os.Args[i+2])
			args = append(args, arg)
			args = append(args, os.Args[i+1]+"="+os.Args[i+2])
			i += 2
			continue
		}
		if (strings.HasPrefix(arg, "-X=") || strings.HasPrefix(arg, "--X=")) && i+1 < len(os.Args) && strings.Count(arg, "=") == 1 {
			fmt.Fprintf(os.Stderr, "link: warning: option %s %s may not work in future releases; use %s=%s\n",
				arg, os.Args[i+1],
				arg, os.Args[i+1])
			args = append(args, arg+"="+os.Args[i+1])
			i++
			continue
		}
		args = append(args, arg)
	}
	os.Args = args

	obj.Flagparse(usage)

	startProfile()
	Ctxt.Bso = &Bso
	Ctxt.Debugvlog = int32(Debug['v'])
	if flagShared != 0 {
		if Buildmode == BuildmodeUnset {
			Buildmode = BuildmodeCShared
		} else if Buildmode != BuildmodeCShared {
			Exitf("-shared and -buildmode=%s are incompatible", Buildmode.String())
		}
	}
	if Buildmode == BuildmodeUnset {
		Buildmode = BuildmodeExe
	}

	if Buildmode != BuildmodeShared && flag.NArg() != 1 {
		usage()
	}

	if outfile == "" {
		outfile = "a.out"
		if HEADTYPE == obj.Hwindows {
			outfile += ".exe"
		}
	}

	libinit() // creates outfile

	if HEADTYPE == -1 {
		HEADTYPE = int32(headtype(goos))
	}
	Ctxt.Headtype = int(HEADTYPE)
	if headstring == "" {
		headstring = Headstr(int(HEADTYPE))
	}

	Thearch.Archinit()

	if Linkshared && !Iself {
		Exitf("-linkshared can only be used on elf systems")
	}

	if Debug['v'] != 0 {
		fmt.Fprintf(&Bso, "HEADER = -H%d -T0x%x -D0x%x -R0x%x\n", HEADTYPE, uint64(INITTEXT), uint64(INITDAT), uint32(INITRND))
	}
	Bso.Flush()

	if Buildmode == BuildmodeShared {
		for i := 0; i < flag.NArg(); i++ {
			arg := flag.Arg(i)
			parts := strings.SplitN(arg, "=", 2)
			var pkgpath, file string
			if len(parts) == 1 {
				pkgpath, file = "main", arg
			} else {
				pkgpath, file = parts[0], parts[1]
			}
			pkglistfornote = append(pkglistfornote, pkgpath...)
			pkglistfornote = append(pkglistfornote, '\n')
			addlibpath(Ctxt, "command line", "command line", file, pkgpath, "")
		}
	} else {
		addlibpath(Ctxt, "command line", "command line", flag.Arg(0), "main", "")
	}
	loadlib()

	if Thearch.Thechar == '5' {
		// mark some functions that are only referenced after linker code editing
		if Ctxt.Goarm == 5 {
			mark(Linkrlookup(Ctxt, "_sfloat", 0))
		}
		mark(Linklookup(Ctxt, "runtime.read_tls_fallback", 0))
	}

	checkgo()
	checkstrdata()
	deadcode()
	callgraph()

	doelf()
	if HEADTYPE == obj.Hdarwin {
		domacho()
	}
	dostkcheck()
	if HEADTYPE == obj.Hwindows {
		dope()
	}
	addexport()
	Thearch.Gentext() // trampolines, call stubs, etc.
	textbuildid()
	textaddress()
	pclntab()
	findfunctab()
	symtab()
	dodata()
	address()
	doweak()
	reloc()
	Thearch.Asmb()
	undef()
	hostlink()
	archive()
	if Debug['v'] != 0 {
		fmt.Fprintf(&Bso, "%5.2f cpu time\n", obj.Cputime())
		fmt.Fprintf(&Bso, "%d symbols\n", Ctxt.Nsymbol)
		fmt.Fprintf(&Bso, "%d liveness data\n", liveness)
	}

	Bso.Flush()

	errorexit()
}
Exemple #7
0
func blockcopy(n, ns *gc.Node, osrc, odst, w int64) {
	var noddi gc.Node
	gc.Nodreg(&noddi, gc.Types[gc.Tptr], x86.REG_DI)
	var nodsi gc.Node
	gc.Nodreg(&nodsi, gc.Types[gc.Tptr], x86.REG_SI)

	var nodl gc.Node
	var nodr gc.Node
	if n.Ullman >= ns.Ullman {
		gc.Agenr(n, &nodr, &nodsi)
		if ns.Op == gc.ONAME {
			gc.Gvardef(ns)
		}
		gc.Agenr(ns, &nodl, &noddi)
	} else {
		if ns.Op == gc.ONAME {
			gc.Gvardef(ns)
		}
		gc.Agenr(ns, &nodl, &noddi)
		gc.Agenr(n, &nodr, &nodsi)
	}

	if nodl.Reg != x86.REG_DI {
		gmove(&nodl, &noddi)
	}
	if nodr.Reg != x86.REG_SI {
		gmove(&nodr, &nodsi)
	}
	gc.Regfree(&nodl)
	gc.Regfree(&nodr)

	c := w % 8 // bytes
	q := w / 8 // quads

	var oldcx gc.Node
	var cx gc.Node
	savex(x86.REG_CX, &cx, &oldcx, nil, gc.Types[gc.TINT64])

	// if we are copying forward on the stack and
	// the src and dst overlap, then reverse direction
	if osrc < odst && odst < osrc+w {
		// reverse direction
		gins(x86.ASTD, nil, nil) // set direction flag
		if c > 0 {
			gconreg(addptr, w-1, x86.REG_SI)
			gconreg(addptr, w-1, x86.REG_DI)

			gconreg(movptr, c, x86.REG_CX)
			gins(x86.AREP, nil, nil)   // repeat
			gins(x86.AMOVSB, nil, nil) // MOVB *(SI)-,*(DI)-
		}

		if q > 0 {
			if c > 0 {
				gconreg(addptr, -7, x86.REG_SI)
				gconreg(addptr, -7, x86.REG_DI)
			} else {
				gconreg(addptr, w-8, x86.REG_SI)
				gconreg(addptr, w-8, x86.REG_DI)
			}

			gconreg(movptr, q, x86.REG_CX)
			gins(x86.AREP, nil, nil)   // repeat
			gins(x86.AMOVSQ, nil, nil) // MOVQ *(SI)-,*(DI)-
		}

		// we leave with the flag clear
		gins(x86.ACLD, nil, nil)
	} else {
		// normal direction
		if q > 128 || (gc.Nacl && q >= 4) || (obj.Getgoos() == "plan9" && q >= 4) {
			gconreg(movptr, q, x86.REG_CX)
			gins(x86.AREP, nil, nil)   // repeat
			gins(x86.AMOVSQ, nil, nil) // MOVQ *(SI)+,*(DI)+
		} else if q >= 4 {
			var oldx0 gc.Node
			var x0 gc.Node
			savex(x86.REG_X0, &x0, &oldx0, nil, gc.Types[gc.TFLOAT64])

			p := gins(obj.ADUFFCOPY, nil, nil)
			p.To.Type = obj.TYPE_ADDR
			p.To.Sym = gc.Linksym(gc.Pkglookup("duffcopy", gc.Runtimepkg))

			// 64 blocks taking 14 bytes each
			// see ../../../../runtime/mkduff.go
			p.To.Offset = 14 * (64 - q/2)
			restx(&x0, &oldx0)

			if q%2 != 0 {
				gins(x86.AMOVSQ, nil, nil) // MOVQ *(SI)+,*(DI)+
			}
		} else if !gc.Nacl && c == 0 {
			// We don't need the MOVSQ side-effect of updating SI and DI,
			// and issuing a sequence of MOVQs directly is faster.
			nodsi.Op = gc.OINDREG

			noddi.Op = gc.OINDREG
			for q > 0 {
				gmove(&nodsi, &cx) // MOVQ x+(SI),CX
				gmove(&cx, &noddi) // MOVQ CX,x+(DI)
				nodsi.Xoffset += 8
				noddi.Xoffset += 8
				q--
			}
		} else {
			for q > 0 {
				gins(x86.AMOVSQ, nil, nil) // MOVQ *(SI)+,*(DI)+
				q--
			}
		}

		// copy the remaining c bytes
		if w < 4 || c <= 1 || (odst < osrc && osrc < odst+w) {
			for c > 0 {
				gins(x86.AMOVSB, nil, nil) // MOVB *(SI)+,*(DI)+
				c--
			}
		} else if w < 8 || c <= 4 {
			nodsi.Op = gc.OINDREG
			noddi.Op = gc.OINDREG
			cx.Type = gc.Types[gc.TINT32]
			nodsi.Type = gc.Types[gc.TINT32]
			noddi.Type = gc.Types[gc.TINT32]
			if c > 4 {
				nodsi.Xoffset = 0
				noddi.Xoffset = 0
				gmove(&nodsi, &cx)
				gmove(&cx, &noddi)
			}

			nodsi.Xoffset = c - 4
			noddi.Xoffset = c - 4
			gmove(&nodsi, &cx)
			gmove(&cx, &noddi)
		} else {
			nodsi.Op = gc.OINDREG
			noddi.Op = gc.OINDREG
			cx.Type = gc.Types[gc.TINT64]
			nodsi.Type = gc.Types[gc.TINT64]
			noddi.Type = gc.Types[gc.TINT64]
			nodsi.Xoffset = c - 8
			noddi.Xoffset = c - 8
			gmove(&nodsi, &cx)
			gmove(&cx, &noddi)
		}
	}

	restx(&cx, &oldcx)
}