BLT	tail		// Jump to the small string case
	SRD	$5, R4, R20
	MOVD	R20, CTR
	MOVD	$16, R21
	XXLXOR	V4, V4, V4
	XXLXOR	V5, V5, V5

	PCALIGN	$16
cmploop:
	LXVD2X	(R0)(R3), V0	// Count 32B per loop with two vector accumulators.
	LXVD2X	(R21)(R3), V2
	VCMPEQUB V2, V1, V2
	VCMPEQUB V0, V1, V0
	VPOPCNTD V2, V2		// A match is 0xFF or 0. Count the bits into doubleword buckets.
	VPOPCNTD V0, V0

	MOVD	$0x0c0, R_x0c0
	MOVD	$0x0d0, R_x0d0
	MOVD	$0x0e0, R_x0e0
	MOVD	$0x0f0, R_x0f0
	MOVD	$0x100, R_x100
	MOVD	$0x110, R_x110

loop:
	MOVD	TBL_STRT, TBL
	LVX	(TBL)(R_x000), KI

	LXVD2X	(INP)(R_x000), V8 // load v8 in advance

	// Offload to VSR24-31 (aka FPR24-31)
	XXLOR	V0, V0, VS24
	XXLOR	V1, V1, VS25
	XXLOR	V2, V2, VS26
	XXLOR	V3, V3, VS27
	XXLOR	V4, V4, VS28
	XXLOR	V5, V5, VS29

	MOVD	$0x0c0, R_x0c0
	MOVD	$0x0d0, R_x0d0
	MOVD	$0x0e0, R_x0e0
	MOVD	$0x0f0, R_x0f0
	MOVD	$0x100, R_x100
	MOVD	$0x110, R_x110

loop:
	MOVD	TBL_STRT, TBL
	LVX	(TBL)(R_x000), KI

	LXVD2X	(INP)(R_x000), V8 // load v8 in advance

	// Offload to VSR24-31 (aka FPR24-31)
	XXLOR	V0, V0, VS24
	XXLOR	V1, V1, VS25
	XXLOR	V2, V2, VS26
	XXLOR	V3, V3, VS27
	XXLOR	V4, V4, VS28
	XXLOR	V5, V5, VS29

	"VNCIPH",
	"VNCIPHER",
	"VNCIPHERLAST",
	"VSBOX",
	"VSHASIGMA",
	"VSHASIGMAW",
	"VSHASIGMAD",
	"VMRGEW",
	"VMRGOW",
	"VCLZLSBB",
	"VCTZLSBB",
	"LXV",
	"LXVL",
	"LXVLL",
	"LXVD2X",
	"LXVW4X",
	"LXVH8X",
	"LXVB16X",
	"LXVX",
	"LXVDSX",
	"STXV",
	"STXVL",
	"STXVLL",
	"STXVD2X",
	"STXVW4X",
	"STXVH8X",
	"STXVB16X",
	"STXVX",
	"LXSDX",
	"STXSDX",

		if args[1] == "0" {
			return op + " (" + args[2] + ")," + args[0]
		}
		return op + " (" + args[2] + ")(" + args[1] + ")," + args[0]

	case LXVX, LXVD2X, LXVW4X, LXVH8X, LXVB16X, LVX, LVXL, LVSR, LVSL, LVEBX, LVEHX, LVEWX, LXSDX, LXSIWAX:
		return op + " (" + args[2] + ")(" + args[1] + ")," + args[0]

	case LXV:
		return op + " " + args[1] + "," + args[0]

// Unrolled index2to16 loop by 4 on ppc64le/power9
// Work is still needed for a big endian
// implementation on power9.

//go:build ppc64 || ppc64le

#include "go_asm.h"
#include "textflag.h"

// Needed to swap LXVD2X loads to the correct
// byte order to work on POWER8.

#ifdef GOARCH_ppc64
DATA byteswap<>+0(SB)/8, $0x0001020304050607
DATA byteswap<>+8(SB)/8, $0x08090a0b0c0d0e0f
#else
DATA byteswap<>+0(SB)/8, $0x0706050403020100

	VSPLTB  $7, V6, V4
	MTVSRD  R5, VS39        // ŝ
	VSPLTB  $7, V7, V2
	ADD     $-2, R4, R16
	PCALIGN $16
loopback:
	ADD     $-1, R8, R10
	SLD     $3, R10
	LXVD2X  (R6)(R10), VS32 // load x[i-1], x[i]
	SLD     $3, R8, R12
	LXVD2X  (R6)(R12), VS33 // load x[i], x[i+1]

	VSRD    V0, V4, V3      // x[i-1]>>s, x[i]>>s
	VSLD    V1, V2, V5      // x[i]<<ŝ, x[i+1]<<ŝ

	VSHASIGMAW $1, $15, V1, V2      // 10418e82
	VSHASIGMAD $2, V1, $15, V2      // 104196c2
	VSHASIGMAD $2, $15, V1, V2      // 104196c2

	LXVD2X (R3)(R4), VS1            // 7c241e98
	LXVD2X (R3)(R0), VS1            // 7c201e98
	LXVD2X (R3), VS1                // 7c201e98
	LXVDSX (R3)(R4), VS1            // 7c241a98
	LXVDSX (R3)(R0), VS1            // 7c201a98
	LXVDSX (R3), VS1                // 7c201a98

			unsafePoint:    true,
		},

		// R31 is temp register
		// Loop code:
		//	MOVD len/32,R31		set up loop ctr
		//	MOVD R31,CTR
		//	MOVD $16,R31		index register
		// loop:
		//	LXVD2X (R0)(R4),VS32
		//	LXVD2X (R31)(R4),VS33
		//	ADD  R4,$32          increment src
		//	STXVD2X VS32,(R0)(R3)
		//	STXVD2X VS33,(R31)(R3)
		//	ADD  R3,$32          increment dst
		//	BC 16,0,loop         branch ctr

		// number of moves are generated based on the
		// size.
		// When moving >= 64 bytes a loop is used
		//	MOVD len/32,REG_TMP
		//	MOVD REG_TMP,CTR
		//	MOVD $16,REG_TMP
		// top:
		//	LXVD2X (R0)(R21),VS32
		//	LXVD2X (R31)(R21),VS33
		//	ADD $32,R21
		//	STXVD2X VS32,(R0)(R20)
		//	STXVD2X VS33,(R31)(R20)
		//	ADD $32,R20
		//	BC 16,0,top
		// Bytes not moved by this loop are moved