IMUL3Q $0x13, AX, AX
	MULQ   24(BX)
	ADDQ   AX, R9
	ADCQ   DX, R8

	// r1 += 19×a4×b2
	MOVQ   32(CX), AX
	IMUL3Q $0x13, AX, AX
	MULQ   16(BX)
	ADDQ   AX, R9
	ADCQ   DX, R8

	// r2 = a0×b2
	MOVQ (CX), AX
	MULQ 16(BX)
	MOVQ AX, R11
	MOVQ DX, R10

	// r2 += a1×b1
	MOVQ 8(CX), AX
	MULQ 8(BX)
	ADDQ AX, R11
	ADCQ DX, R10

	// r2 += a2×b0
	MOVQ 16(CX), AX
	MULQ (BX)
	ADDQ AX, R11

	ADCQ $1, h2;      \
	LEAQ 16(msg), msg

#define POLY1305_MUL(h0, h1, h2, r0, r1, t0, t1, t2, t3) \
	MOVQ  r0, AX;                  \
	MULQ  h0;                      \
	MOVQ  AX, t0;                  \
	MOVQ  DX, t1;                  \
	MOVQ  r0, AX;                  \
	MULQ  h1;                      \
	ADDQ  AX, t1;                  \
	ADCQ  $0, DX;                  \
	MOVQ  r0, t2;                  \

	MOVQ (0*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (0*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (1*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (1*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (2*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (2*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (3*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX

Regexps are always matched from the start of the instructions line.
This means, for example, that the "MULQ" regexp is equivalent to
"^MULQ" (^ representing the start of the line), and it will NOT match
the following assembly line:

  IMULQ	$99, AX

To force a match at any point of the line, ".*MULQ" should be used.

For the same reason, a negative regexp like -"memmove" is not enough

	carry := GP64()
	XORQ(carry, carry) // zero out carry

	for i := 0; i < bits/64; i++ {
		Comment("Iteration " + strconv.Itoa(i))
		hi, lo := RDX, RAX // implicit MULQ inputs and outputs
		MOVQ(x.Offset(i*8), lo)
		MULQ(y)
		ADDQ(z.Offset(i*8), lo)
		ADCQ(Imm(0), hi)
		ADDQ(carry, lo)
		ADCQ(Imm(0), hi)
		MOVQ(hi, carry)
		MOVQ(lo, z.Offset(i*8))
	}

	Store(carry, ReturnIndex(0))
	RET()

// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

// Issue 16733: don't fold constant factors into a multiply
// beyond the capacity of a MULQ instruction (32 bits).

package p

func f(n int64) int64 {
	n *= 1000000
	n *= 1000000
	return n

	SUBQ	DI, AX
	IMULQ	$100, AX

	// generated code for
	//	func f(x uint64) (uint64, uint64) { return x/1000000000, x%1000000000 }
	// adapted to reduce duplication
	MOVQ	AX, CX
	MOVQ	$1360296554856532783, AX
	MULQ	CX
	ADDQ	CX, DX
	RCRQ	$1, DX
	SHRQ	$29, DX
	MOVQ	DX, sec+0(FP)
	IMULQ	$1000000000, DX
	SUBQ	DX, CX
	MOVL	CX, nsec+8(FP)

		Load(aX, RAX)
	case 2:
		Comment(fmt.Sprintf("%s = 2×%s×%s", r, aX, bX))
		Load(aX, RAX)
		SHLQ(Imm(1), RAX)
	default:
		panic("unsupported i value")
	}
	MULQ(mustAddr(bX)) // RDX, RAX = RAX * bX
	MOVQ(RAX, r.lo)
	MOVQ(RDX, r.hi)
}

// addMul64 sets r to r + i * aX * bX.
func addMul64(r uint128, i uint64, aX, bX namedComponent) {
	switch i {
	case 1:

	MOVQ	0(SP), AX

	// generated code for
	//	func f(x uint64) (uint64, uint64) { return x/1000000000, x%1000000000 }
	// adapted to reduce duplication
	MOVQ	AX, CX
	MOVQ	$1360296554856532783, AX
	MULQ	CX
	ADDQ	CX, DX
	RCRQ	$1, DX
	SHRQ	$29, DX
	MOVQ	DX, sec+0(FP)
	IMULQ	$1000000000, DX
	SUBQ	DX, CX
	MOVL	CX, nsec+8(FP)
	RET

TEXT runtime·notify(SB),NOSPLIT,$0
	MOVQ	$28, BP
	SYSCALL