opset(AVSHASIGMAW, r0)
			opset(AVSHASIGMAD, r0)

		case ALXVD2X: /* lxvd2x, lxvdsx, lxvw4x, lxvh8x, lxvb16x */
			opset(ALXVDSX, r0)
			opset(ALXVW4X, r0)
			opset(ALXVH8X, r0)
			opset(ALXVB16X, r0)

		case ALXV: /* lxv */
			opset(ALXV, r0)

		case ALXVL: /* lxvl, lxvll, lxvx */
			opset(ALXVLL, r0)
			opset(ALXVX, r0)

// 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

			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