{name: "ANDQ", argLength: 2, reg: gp21, asm: "ANDQ", commutative: true, resultInArg0: true, clobberFlags: true},                                                 // arg0 & arg1

	CMPB ·support_adx(SB), $1
	JEQ adx
	MOVQ z+0(FP), R10
	MOVQ x+24(FP), R8
	MOVQ y+48(FP), R9
	MOVQ z_len+8(FP), R11
	MOVQ $0, BX		// i = 0
	MOVQ $0, CX		// c = 0
	MOVQ R11, R12
	ANDQ $-2, R12
	CMPQ R11, $2
	JAE A6
	JMP E6

A6:
	MOVQ (R8)(BX*8), AX
	MULQ R9
	ADDQ (R10)(BX*8), AX
	ADCQ $0, DX
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ DX, CX
	MOVQ AX, (R10)(BX*8)

	MOVQ 	AX, ret+16(FP)
	RET

// func And64(addr *uint64, v uint64) old uint64
TEXT ·And64(SB), NOSPLIT, $0-24
	MOVQ	ptr+0(FP), BX
	MOVQ	val+8(FP), CX
casloop:
	MOVQ 	CX, DX
	MOVQ	(BX), AX
	ANDQ	AX, DX
	LOCK
	CMPXCHGQ	DX, (BX)
	JNZ casloop
	MOVQ 	AX, ret+16(FP)
	RET

// func Anduintptr(addr *uintptr, v uintptr) old uintptr
TEXT ·Anduintptr(SB), NOSPLIT, $0-24
	JMP	·And64(SB)

	MOVL crc+0(FP), AX  // CRC value
	MOVQ p+8(FP), SI  // data pointer
	MOVQ p_len+16(FP), CX  // len(p)

	// If there are fewer than 8 bytes to process, skip alignment.
	CMPQ CX, $8
	JL less_than_8

	MOVQ SI, BX
	ANDQ $7, BX
	JZ aligned

	// Process the first few bytes to 8-byte align the input.

	// BX = 8 - BX. We need to process this many bytes to align.
	SUBQ $1, BX
	XORQ $7, BX

	BTQ $0, BX
	JNC align_2

	lo = r.lo
	SHLQ(Imm(64-51), r.lo, r.hi)
	r.lo, r.hi = nil, nil // make sure the uint128 is unusable
	return
}

// maskAndAdd sets r = r&mask + c*i.
func maskAndAdd(r, mask, c GPVirtual, i uint64) {
	ANDQ(mask, r)
	if i != 1 {
		IMUL3Q(Imm(i), c, c)
	}
	ADDQ(c, r)
}

func mustAddr(c Component) Op {
	b, err := c.Resolve()
	if err != nil {
		panic(err)
	}
	return b.Addr

	ADDQ    $168, SP
	RET

TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
	MOVQ	fn+0(FP),    AX
	MOVL	sig+8(FP),   DI
	MOVQ	info+16(FP), SI
	MOVQ	ctx+24(FP),  DX
	MOVQ	SP, BX		// callee-saved
	ANDQ	$~15, SP	// alignment for x86_64 ABI
	CALL	AX
	MOVQ	BX, SP
	RET

// Called from runtime·usleep (Go). Can be called on Go stack, on OS stack,
// can also be called in cgo callback path without a g->m.

	MOVQ	_cgo_sys_thread_create(SB), AX
	TESTQ	AX, AX
	JZ	nocgo

	// We're calling back to C.
	// Align stack per ELF ABI requirements.
	MOVQ	SP, BX  // Callee-save in C ABI
	ANDQ	$~15, SP
	MOVQ	$_rt0_amd64_lib_go(SB), DI
	MOVQ	$0, SI
	CALL	AX
	MOVQ	BX, SP
	JMP	restore

nocgo:
	ADJSP	$16
	MOVQ	$0x800000, 0(SP)		// stacksize
	MOVQ	$_rt0_amd64_lib_go(SB), AX
	MOVQ	AX, 8(SP)			// fn

func bitoff64(a, b uint64) (n uint64) {
	// amd64:"BTRQ"
	n += b &^ (1 << (a & 63))

	// amd64:"BTRQ\t[$]63"
	n += a &^ (1 << 63)

	// amd64:"BTRQ\t[$]60"
	n += a &^ (1 << 60)

	// amd64:"ANDQ\t[$]-2"
	n += a &^ (1 << 0)

	return n
}

func bitcompl64(a, b uint64) (n uint64) {
	// amd64:"BTCQ"
	n += b ^ (1 << (a & 63))

	// amd64:"BTCQ\t[$]63"
	n += a ^ (1 << 63)

	MOVL	$0xf1, 0xf1		// crash
	RET

TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
	MOVQ	fn+0(FP),    AX
	MOVL	sig+8(FP),   DI
	MOVQ	info+16(FP), SI
	MOVQ	ctx+24(FP),  DX
	MOVQ	SP, BX		// callee-saved
	ANDQ	$~15, SP	// alignment for x86_64 ABI
	CALL	AX
	MOVQ	BX, SP
	RET

// Called using C ABI.
TEXT runtime·sigtramp(SB),NOSPLIT|TOPFRAME|NOFRAME,$0
	// Transition from C ABI to Go ABI.
	PUSH_REGS_HOST_TO_ABI0()

#define polyMulStage3 IMULQ acc2, t3; ADDQ acc0, t2; ADCQ DX, t3
#define polyMulReduceStage MOVQ t0, acc0; MOVQ t1, acc1; MOVQ t2, acc2; ANDQ $3, acc2; MOVQ t2, t0; ANDQ $-4, t0; MOVQ t3, t1; SHRQ $2, t3, t2; SHRQ $2, t3; ADDQ t0, acc0; ADCQ t1, acc1; ADCQ $0, acc2; ADDQ t2, acc0; ADCQ t3, acc1; ADCQ $0, acc2