switch op {
			case "MOVB", "MOVBU":
				src = 1
			case "MOVH", "MOVHU":
				src = 2
			case "MOVW", "MOVWU", "MOVF":
				src = 4
			case "MOVV", "MOVD":
				src = 8
			}
		case "s390x":
			switch op {
			case "MOVB", "MOVBZ":
				src = 1
			case "MOVH", "MOVHZ":
				src = 2
			case "MOVW", "MOVWZ", "FMOVS":
				src = 4
			case "MOVD", "FMOVD":
				src = 8
			}
		}
	}

	MOVD	$runtime·m0(SB), R0

	// save m->g0 = g0
	MOVD	g, m_g0(R0)
	// save m0 to g0->m
	MOVD	R0, g_m(g)

	BL	runtime·check(SB)

#ifdef GOOS_windows
	BL	runtime·wintls(SB)
#endif

	MOVW	8(RSP), R0	// copy argc
	MOVW	R0, -8(RSP)
	MOVD	16(RSP), R0		// copy argv
	MOVD	R0, 0(RSP)
	BL	runtime·args(SB)
	BL	runtime·osinit(SB)
	BL	runtime·schedinit(SB)

	// create a new goroutine to start program

move_1or2:
	MOVB	(SI), AX
	MOVB	-1(SI)(BX*1), CX
	MOVB	AX, (DI)
	MOVB	CX, -1(DI)(BX*1)
	RET
move_0:
	RET
move_4:
	MOVL	(SI), AX
	MOVL	AX, (DI)
	RET
move_3:
	MOVW	(SI), AX
	MOVB	2(SI), CX
	MOVW	AX, (DI)
	MOVB	CX, 2(DI)
	RET
move_5through7:
	MOVL	(SI), AX
	MOVL	-4(SI)(BX*1), CX
	MOVL	AX, (DI)
	MOVL	CX, -4(DI)(BX*1)
	RET
move_8:

	// ppc64x:"ROTL"
	// s390x:"RISBGZ\t[$]0, [$]63, [$]37, "
	// wasm:"I64Rotl"
	return bits.RotateLeft64(n, 37)
}

func RotateLeft32(n uint32) uint32 {
	// amd64:"ROLL" 386:"ROLL"
	// arm:`MOVW\tR[0-9]+@>23`
	// arm64:"RORW"
	// ppc64x:"ROTLW"
	// s390x:"RLL"
	// wasm:"I32Rotl"
	return bits.RotateLeft32(n, 9)
}

func RotateLeft16(n uint16, s int) uint16 {
	// amd64:"ROLW" 386:"ROLW"

	CMP	$128, autLen
	BLT	startSinglesLoop
	B	octetsLoop

dataTLS:
	ADD	$14*16, pTbl
	VLD1.P	(pTbl), [T1.B16, T2.B16]
	VEOR	B0.B16, B0.B16, B0.B16

	MOVD	(aut), H0
	VMOV	H0, B0.D[0]
	MOVW	8(aut), H0
	VMOV	H0, B0.S[2]
	MOVB	12(aut), H0
	VMOV	H0, B0.B[12]

	MOVD	$0, autLen
	B	dataMul

octetsLoop:
		CMP	$128, autLen
		BLT	startSinglesLoop
		SUB	$128, autLen

</p>

<p>
Instruction modifiers are appended to the instruction following a period.
The only modifiers are <code>P</code> (postincrement) and <code>W</code>
(preincrement):
<code>MOVW.P</code>, <code>MOVW.W</code>
</p>

<p>
Addressing modes:
</p>

<ul>

<li>
<code>R0-&gt;16</code>
<br>
<code>R0&gt;&gt;16</code>
<br>
<code>R0&lt;&lt;16</code>
<br>

	"MOVOU",
	"MOVQ",
	"MOVQL",
	"MOVQOZX",
	"MOVSB",
	"MOVSD",
	"MOVSHDUP",
	"MOVSL",
	"MOVSLDUP",
	"MOVSQ",
	"MOVSS",
	"MOVSW",
	"MOVSWW",
	"MOVUPD",
	"MOVUPS",
	"MOVW",
	"MOVWLSX",
	"MOVWLZX",
	"MOVWQSX",
	"MOVWQZX",
	"MOVZWW",
	"MPSADBW",
	"MULB",
	"MULL",
	"MULPD",
	"MULPS",
	"MULQ",
	"MULSD",
	"MULSS",
	"MULW",
	"MULXL",

	// store entry number if the kernel allocated it
	CMPL	CX, $-1
	JNE	2(PC)
	MOVL	AX, runtime·tls_entry_number(SB)

	// compute segment selector - (entry*8+3)
	SHLL	$3, AX
	ADDL	$3, AX
	MOVW	AX, GS

	RET

TEXT runtime·osyield(SB),NOSPLIT,$0
	MOVL	$SYS_sched_yield, AX
	INVOKE_SYSCALL
	RET

TEXT runtime·sched_getaffinity(SB),NOSPLIT,$0
	MOVL	$SYS_sched_getaffinity, AX

	JALR	RA, (X5)

	// Restore g, stack pointer. X10 is return value.
	MOV	0(X2), g
	CALL	runtime·save_g(SB)
	MOV	(g_stack+stack_hi)(g), X5
	MOV	8(X2), X6
	SUB	X6, X5, X6
	MOV	X6, X2

	MOVW	X10, ret+16(FP)
	RET

// func asminit()
TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
	RET

// reflectcall: call a function with the given argument list

	XOR	R14, R15, R14
	MOVD	R14, 0(BLK_OUT)
	ADD	$8, R16
	ADD	$-8, IN_LEN
next4:
	CMP	IN_LEN, $4
	BLT	next2
	MOVWZ	(BLK_INP)(R16), R15
	MOVWZ	(MASK_PTR)(R16), R14
	XOR	R14, R15, R14
	MOVW	R14, (R16)(BLK_OUT)
	ADD	$4, R16
	ADD	$-4, IN_LEN
next2:
	CMP	IN_LEN, $2
	BLT	next1
	MOVHZ	(BLK_INP)(R16), R15
	MOVHZ	(MASK_PTR)(R16), R14
	XOR	R14, R15, R14
	MOVH	R14, (R16)(BLK_OUT)