DENBCDQCC:      "denbcdq.",
	DIEX:           "diex",
	DIEXCC:         "diex.",
	DIEXQCC:        "diexq.",
	DIEXQ:          "diexq",
	DMUL:           "dmul",
	DMULCC:         "dmul.",
	DMULQ:          "dmulq",
	DMULQCC:        "dmulq.",
	DQUA:           "dqua",
	DQUACC:         "dqua.",
	DQUAI:          "dquai",
	DQUAICC:        "dquai.",
	DQUAIQ:         "dquaiq",

#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;                  \
	IMULQ h2, t2;                  \
	ADDQ  DX, t2;                  \

	MOVQ	$_INTERRUPT_TIME, DI
	MOVQ	time_lo(DI), AX
	IMULQ	$100, AX
	MOVQ	AX, mono+16(FP)

	MOVQ	$_SYSTEM_TIME, DI
	MOVQ	time_lo(DI), AX
	MOVQ	$116444736000000000, DI
	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

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

//    Multiplication    //
// -------------------- //

func Pow2Muls(n1, n2 int) (int, int) {
	// amd64:"SHLQ\t[$]5",-"IMULQ"
	// 386:"SHLL\t[$]5",-"IMULL"
	// arm:"SLL\t[$]5",-"MUL"
	// arm64:"LSL\t[$]5",-"MUL"
	// ppc64x:"SLD\t[$]5",-"MUL"
	a := n1 * 32

	// amd64:"SHLQ\t[$]6",-"IMULQ"
	// 386:"SHLL\t[$]6",-"IMULL"
	// arm:"SLL\t[$]6",-"MUL"
	// arm64:`NEG\sR[0-9]+<<6,\sR[0-9]+`,-`LSL`,-`MUL`

	// 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
	MOVL	AX, ret+8(FP)
	RET

	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

	XORQ DI, DI

	// Iteration 0
	MOVQ (BX), AX
	MULQ SI
	ADDQ (CX), AX
	ADCQ $0x00, DX
	ADDQ DI, AX
	ADCQ $0x00, DX
	MOVQ DX, DI
	MOVQ AX, (CX)

	// Iteration 1
	MOVQ 8(BX), AX
	MULQ SI
	ADDQ 8(CX), AX
	ADCQ $0x00, DX
	ADDQ DI, AX
	ADCQ $0x00, DX
	MOVQ DX, DI
	MOVQ AX, 8(CX)

	// Iteration 2
	MOVQ 16(BX), AX
	MULQ SI
	ADDQ 16(CX), AX
	ADCQ $0x00, DX
	ADDQ DI, AX

	MOVQ (8*0)(x_ptr), t0

	MOVQ (8*1)(x_ptr), AX
	MULQ t0
	MOVQ AX, acc1
	MOVQ DX, acc2

	MOVQ (8*2)(x_ptr), AX
	MULQ t0
	ADDQ AX, acc2
	ADCQ $0, DX
	MOVQ DX, acc3

	MOVQ (8*3)(x_ptr), AX
	MULQ t0
	ADDQ AX, acc3
	ADCQ $0, DX
	MOVQ DX, acc4
	// y[2:] * y[1]
	MOVQ (8*1)(x_ptr), t0

	MOVQ (8*2)(x_ptr), AX
	MULQ t0
	ADDQ AX, acc3
	ADCQ $0, DX
	MOVQ DX, t1

	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