#ifndef GOPPC64_power9
#define ADJUST_FOR_CNTLZW -16
#else
#define ADJUST_FOR_CNTLZW 0
#endif

// Now, find the index of the 16B vector the match was discovered in. If CNTLZW is used
// to determine the offset into the 16B vector, it will overcount by 16. Account for it here.
foundat3:
	SUB	R3,R8,R3
	ADD	$48+ADJUST_FOR_CNTLZW,R3
	BR	vfound
foundat2:
	SUB	R3,R8,R3

	VLD1.P (R6)(R11), [V31.D1]        <=>      ld1 {v31.1d}, [x6], x11
	VFMLA V29.S2, V20.S2, V14.S2      <=>      fmla v14.2s, v20.2s, v29.2s
	AESD V22.B16, V19.B16             <=>      aesd v19.16b, v22.16b
	SCVTFWS R3, F16                   <=>      scvtf s17, w6

6. Align directive

Go asm supports the PCALIGN directive, which indicates that the next instruction should be aligned

	VCMEQ	V0.B16, V2.B16, V4.B16
	// Clear the higher 7 bits
	VAND	V5.B16, V3.B16, V3.B16
	VAND	V5.B16, V4.B16, V4.B16
	// Count lanes match the requested byte
	VADDP	V4.B16, V3.B16, V6.B16 // 32B->16B
	VUADDLV	V6.B16, V7
	// Accumulate the count in low 64-bit element of V8 when inside the loop
	VADD	V7, V8
	BNE	chunk_loop
	VMOV	V8.D[0], R6
	ADD	R6, R11, R11
	CBZ	R2, done
tail:

	ADD	R14, R18, R18
	ADD	$16, R3, R3
	ANDCC	$15, R4, R4

small_tail_p10:
	SLD	$56, R4, R6
	LXVLL	R3, R6, V0
	VCMPEQUB V0, V1, V0
	VCLRRB	V0, R4, V0	// If <16B being compared, clear matches of the 16-R4 bytes.
	VCNTMBB	V0, $1, R14	// Sum the value of bit 0 of each byte of the compare into R14.
	SRD	$56, R14, R14	// The result of VCNTMBB is shifted. Unshift it.
	ADD	R14, R18, R3
	RET

	FMOVD	(offset+8*14)(R1), F28 \
	FMOVD	(offset+8*15)(R1), F29 \
	FMOVD	(offset+8*16)(R1), F30 \
	FMOVD	(offset+8*17)(R1), F31

// Save and restore VR20-31 (aka VSR56-63). These
// macros must point to a 16B aligned offset.
#define SAVE_VR_SIZE (12*16)
#define SAVE_VR(offset, rtmp)         \
	MOVD	$(offset+16*0), rtmp  \
	STVX	V20, (rtmp)(R1)       \
	MOVD	$(offset+16*1), rtmp  \
	STVX	V21, (rtmp)(R1)       \

			"a\tb\tc\td\n" +
			"a\tb\tc\td\te\n",
	},

	{
		"16b",
		100, 100, 0, '\t', DiscardEmptyColumns,
		"a\vb\v\vd\n" +
			"a\vb\v\vd\ve\n" +
			"a\n" +
			"a\vb\vc\vd\n" +
			"a\vb\vc\vd\ve\n",

		"a\tb\td\n" +
			"a\tb\td\te\n" +
			"a\n" +
			"a\tb\tc\td\n" +
			"a\tb\tc\td\te\n",
	},

	{
		"16b debug",
		100, 100, 0, '\t', DiscardEmptyColumns | Debug,
		"a\vb\v\vd\n" +

	PCALIGN $16
cmp64_tail_gt16: // 17 - 32B
	LXVD2X	(R0)(R5),V3
	LXVD2X	(R0)(R6),V4
	VCMPEQUDCC	V3,V4,V1
	BGE	CR6,different

	BR cmp64_tail_gt0

	PCALIGN $16
cmp64_tail_gt0: // 1 - 16B
	LXVD2X	(R5)(R9),V3
	LXVD2X	(R6)(R9),V4
	VCMPEQUDCC	V3,V4,V1
	BGE	CR6,different

	RET

	PCALIGN $16
cmp32:	// 32 - 63B
	ANDCC	$31,R9,R9

	LXVD2X	(R0)(R5),V3
	LXVD2X	(R0)(R6),V4

			if target.IsAIX() && rt == objabi.R_POWER_TLS_LE {
				// Fixup val, an addis/addi pair of instructions, which generate a 32b displacement
				// from the threadpointer (R13), into a 16b relocation. XCOFF only supports 16b
				// TLS LE relocations. Likewise, verify this is an addis/addi sequence.
				const expectedOpcodes = 0x3C00000038000000
				const expectedOpmasks = 0xFC000000FC000000

	MOVD	0(R10), R11
	MOVD	g_m(R11), R3
	MOVD	m_p(R3), R3
	MOVD	p_raceprocctx(R3), R3
	MOVD	R3, (R4)
	RET

rest:
	// Save registers according to the host PPC64 ABI
	// and reserve 16B for argument storage.
	STACK_AND_SAVE_HOST_TO_GO_ABI(16)

	// Load g, and switch to g0 if not already on it.
	MOVD	runtime·tls_g(SB), R10
	MOVD	0(R10), g

	MOVD	g_m(g), R7
	MOVD	m_g0(R7), R8
	CMP	g, R8

		{[]byte(x16pgm), "align=0x20", false},   // Increased alignment needed
		{[]byte(x32pgm), "align=0x40", false},   // Worst case alignment needed
		{[]byte(x64pgm), "align=0x0", true},     // 0 aligned is default (16B) alignment
		{[]byte(x64pgmA64), "align=0x40", true}, // extra alignment + nop
		{[]byte(x64pgmA32), "align=0x20", true}, // extra alignment + nop
	}

	for _, pgm := range pgms {
		tmpfile := filepath.Join(dir, "x.s")