r += v
	}
	return r
}

func eq(v1, v2 int) {
	if v1 != v2 {
		panic(0)
	}
}

func main() {
	a := []int{1, 2, 3}
	t1 := T1(4)
	t2 := &T2{4}

	eq(t1.Sum(a, 5), 15)
	eq(t2.Sum(a, 6), 16)

	eq(T1.Sum(t1, a, 7), 17)
	eq((*T2).Sum(t2, a, 8), 18)

	f1 := T1.Sum
	eq(f1(t1, a, 9), 19)
	f2 := (*T2).Sum
	eq(f2(t2, a, 10), 20)

	eq(I1.Sum(t1, a, 11), 21)
	eq(I1.Sum(t2, a, 12), 22)

	REPLREG(R13, X15)
	PADDD X0, X4
	PADDD X1, X5
	PADDD X2, X6
	PADDD X3, X7
	PADDD X12, X8
	PADDD X13, X9
	PADDD X14, X10
	PADDD X15, X11
	MOVOU X4, (4*16)(BX)
	MOVOU X5, (5*16)(BX)
	MOVOU X6, (6*16)(BX)
	MOVOU X7, (7*16)(BX)
	MOVOU X8, (8*16)(BX)
	MOVOU X9, (9*16)(BX)
	MOVOU X10, (10*16)(BX)
	MOVOU X11, (11*16)(BX)

	MOVL $0, AX
	MOVQ AX, X15 // must be 0 on return

	RET

	MOVOU (3*16)(inp), D3; PXOR D3, D0; MOVOU D0, (3*16)(oup)
	MOVOU (4*16)(inp), D0; PXOR D0, A1; MOVOU A1, (4*16)(oup)
	MOVOU (5*16)(inp), D0; PXOR D0, B1; MOVOU B1, (5*16)(oup)
	MOVOU (6*16)(inp), D0; PXOR D0, C1; MOVOU C1, (6*16)(oup)
	MOVOU (7*16)(inp), D0; PXOR D0, D1; MOVOU D1, (7*16)(oup)
	MOVOU (8*16)(inp), D0; PXOR D0, A2; MOVOU A2, (8*16)(oup)
	MOVOU (9*16)(inp), D0; PXOR D0, B2; MOVOU B2, (9*16)(oup)

DATA ·erfcrodataL38<> + 592(SB)/8, $0.500000000000004237e+00
DATA ·erfcrodataL38<> + 600(SB)/8, $0.416666664838056960e-01
DATA ·erfcrodataL38<> + 608(SB)/8, $0.166666666630345592e+00
DATA ·erfcrodataL38<> + 616(SB)/8, $0.138926439368309441e-02
DATA ·erfcrodataL38<> + 624(SB)/8, $0.833349307718286047e-02
DATA ·erfcrodataL38<> + 632(SB)/8, $-.693147180558298714e+00
DATA ·erfcrodataL38<> + 640(SB)/8, $-.164659495826017651e-11

func gbit64(b []byte) (uint64, []byte) {
	lo := uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
	hi := uint32(b[4]) | uint32(b[5])<<8 | uint32(b[6])<<16 | uint32(b[7])<<24
	return uint64(lo) | uint64(hi)<<32, b[8:]
}

// gstring reads a string from b, prefixed with a 16-bit length in little-endian order.

		// on many platforms. See test/codegen/memcombine.go.
		first32 := uint32(p[0]) | uint32(p[1])<<8 | uint32(p[2])<<16 | uint32(p[3])<<24
		second32 := uint32(p[4]) | uint32(p[5])<<8 | uint32(p[6])<<16 | uint32(p[7])<<24
		if (first32|second32)&0x80808080 != 0 {
			// Found a non ASCII byte (>= RuneSelf).
			break
		}
		p = p[8:]
	}
	n := len(p)
	for i := 0; i < n; {
		pi := p[i]

	{"hw.serialno", []_C_int{6, 17}},
	{"hw.setperf", []_C_int{6, 13}},
	{"hw.smt", []_C_int{6, 24}},
	{"hw.usermem", []_C_int{6, 20}},
	{"hw.uuid", []_C_int{6, 18}},
	{"hw.vendor", []_C_int{6, 14}},
	{"hw.version", []_C_int{6, 16}},
	{"kern.allowdt", []_C_int{1, 65}},
	{"kern.allowkmem", []_C_int{1, 52}},
	{"kern.argmax", []_C_int{1, 8}},
	{"kern.audio", []_C_int{1, 84}},
	{"kern.boottime", []_C_int{1, 21}},

	{"hw.serialno", []_C_int{6, 17}},
	{"hw.setperf", []_C_int{6, 13}},
	{"hw.smt", []_C_int{6, 24}},
	{"hw.usermem", []_C_int{6, 20}},
	{"hw.uuid", []_C_int{6, 18}},
	{"hw.vendor", []_C_int{6, 14}},
	{"hw.version", []_C_int{6, 16}},
	{"kern.allowdt", []_C_int{1, 65}},
	{"kern.allowkmem", []_C_int{1, 52}},
	{"kern.argmax", []_C_int{1, 8}},
	{"kern.audio", []_C_int{1, 84}},
	{"kern.boottime", []_C_int{1, 21}},

	{"hw.serialno", []_C_int{6, 17}},
	{"hw.setperf", []_C_int{6, 13}},
	{"hw.smt", []_C_int{6, 24}},
	{"hw.usermem", []_C_int{6, 20}},
	{"hw.uuid", []_C_int{6, 18}},
	{"hw.vendor", []_C_int{6, 14}},
	{"hw.version", []_C_int{6, 16}},
	{"kern.allowdt", []_C_int{1, 65}},
	{"kern.allowkmem", []_C_int{1, 52}},
	{"kern.argmax", []_C_int{1, 8}},
	{"kern.audio", []_C_int{1, 84}},
	{"kern.boottime", []_C_int{1, 21}},

    // codePoint has at most 21 bits
    return ((0xFL << 4) | (codePoint >>> 18))
        | ((0x80L | (0x3F & (codePoint >>> 12))) << 8)
        | ((0x80L | (0x3F & (codePoint >>> 6))) << 16)
        | ((0x80L | (0x3F & codePoint)) << 24);
  }

  private static long charToThreeUtf8Bytes(char c) {
    return ((0x7L << 5) | (c >>> 12))
        | ((0x80 | (0x3F & (c >>> 6))) << 8)