for i := range A {
		x := A[i]
		y := B[i]
		X := T.Bswap64(x) // ERROR "intrinsic substitution for Bswap64"
		Y := T.Bswap64(y) // ERROR "intrinsic substitution for Bswap64"
		if y != X {
			logf("Bswap64(0x%08x) expected 0x%08x but got 0x%08x\n", x, y, X)
		}
		if x != Y {
			logf("Bswap64(0x%08x) expected 0x%08x but got 0x%08x\n", y, x, Y)
		}

		x32 := uint32(X)
		y32 := uint32(Y >> 32)

		if got := sys.TrailingZeros32(x); got != i {
			t.Errorf("TrailingZeros32(%d)=%d, want %d", x, got, i)
		}
	}
}

func TestBswap64(t *testing.T) {
	x := uint64(0x1122334455667788)
	y := sys.Bswap64(x)
	if y != 0x8877665544332211 {
		t.Errorf("Bswap(%x)=%x, want 0x8877665544332211", x, y)
	}
}
func TestBswap32(t *testing.T) {
	x := uint32(0x11223344)
	y := sys.Bswap32(x)
	if y != 0x44332211 {

// Len8 returns the minimum number of bits required to represent x; the result is 0 for x == 0.
func Len8(x uint8) int {
	return int(len8tab[x])
}

// Bswap64 returns its input with byte order reversed
// 0x0102030405060708 -> 0x0807060504030201
func Bswap64(x uint64) uint64 {
	c8 := uint64(0x00ff00ff00ff00ff)
	a := x >> 8 & c8
	b := (x & c8) << 8
	x = a | b
	c16 := uint64(0x0000ffff0000ffff)
	a = x >> 16 & c16

	BigEndian      bool        //
	UseFMA         bool        // Use hardware FMA operation
	unalignedOK    bool        // Unaligned loads/stores are ok
	haveBswap64    bool        // architecture implements Bswap64
	haveBswap32    bool        // architecture implements Bswap32
	haveBswap16    bool        // architecture implements Bswap16
}

type (
	blockRewriter func(*Block) bool
	valueRewriter func(*Value) bool
)

(BitLen64 x) =>
	(Add32 <typ.Int>
		(BitLen32 <typ.Int> (Int64Hi x))
		(BitLen32 <typ.Int>
			(Or32 <typ.UInt32>
				(Int64Lo x)
				(Zeromask (Int64Hi x)))))

(Bswap64 x) =>
	(Int64Make
		(Bswap32 <typ.UInt32> (Int64Lo x))
		(Bswap32 <typ.UInt32> (Int64Hi x)))

(SignExt32to64 x) => (Int64Make (Signmask x) x)
(SignExt16to64 x) => (SignExt32to64 (SignExt16to32 x))

(Bswap(32|16) x:(MOV(W|H)Zloadidx ptr idx      mem)) => @x.Block (MOV(W|H)BRloadidx ptr idx mem)
(Bswap64 x:(MOVDload [off] {sym} ptr mem)) => @x.Block (MOVDBRload (MOVDaddr <ptr.Type> [off] {sym} ptr) mem)
(Bswap64 x:(MOVDloadidx ptr idx      mem)) => @x.Block (MOVDBRloadidx ptr idx mem)

	{name: "BitLen64", argLength: 1},     // Number of bits in arg[0] (returns 0-64)

	{name: "Bswap16", argLength: 1}, // Swap bytes
	{name: "Bswap32", argLength: 1}, // Swap bytes
	{name: "Bswap64", argLength: 1}, // Swap bytes

	{name: "BitRev8", argLength: 1},  // Reverse the bits in arg[0]
	{name: "BitRev16", argLength: 1}, // Reverse the bits in arg[0]

// and leaves it alone elsewhere.
func bswapIfBigEndian(x uintptr) uintptr {
	if goarch.BigEndian {
		if goarch.PtrSize == 8 {
			return uintptr(sys.Bswap64(uint64(x)))
		}
		return uintptr(sys.Bswap32(uint32(x)))
	}
	return x
}

// newUserArenaChunk allocates a user arena chunk, which maps to a single

// Read is little-endian.
func readUintptr(p *byte) uintptr {
	x := *(*uintptr)(unsafe.Pointer(p))
	if goarch.BigEndian {
		if goarch.PtrSize == 8 {
			return uintptr(sys.Bswap64(uint64(x)))
		}
		return uintptr(sys.Bswap32(uint32(x)))
	}
	return x
}

var debugPtrmask struct {
	lock mutex
	data *byte
}

		v2.AddArg(v3)
		v.AddArg2(v0, v2)
		return true
	}
}
func rewriteValuedec64_OpBswap64(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (Bswap64 x)
	// result: (Int64Make (Bswap32 <typ.UInt32> (Int64Lo x)) (Bswap32 <typ.UInt32> (Int64Hi x)))
	for {
		x := v_0
		v.reset(OpInt64Make)
		v0 := b.NewValue0(v.Pos, OpBswap32, typ.UInt32)