"VCMPGTSWCC",
	"VCMPGTSD",
	"VCMPGTSDCC",
	"VCMPNEZB",
	"VCMPNEZBCC",
	"VCMPNEB",
	"VCMPNEBCC",
	"VCMPNEH",
	"VCMPNEHCC",
	"VCMPNEW",
	"VCMPNEWCC",
	"VPERM",
	"VPERMXOR",
	"VPERMR",
	"VBPERMQ",
	"VBPERMD",
	"VSEL",
	"VSPLTB",
	"VSPLTH",
	"VSPLTW",
	"VSPLTISB",
	"VSPLTISH",
	"VSPLTISW",
	"VCIPH",
	"VCIPHER",
	"VCIPHERLAST",
	"VNCIPH",
	"VNCIPHER",

	VCMPNEW V1, V2, V3              // 10611087
	VCMPNEWCC V1, V2, V3            // 10611487
	VPERM V1, V2, V3, V4            // 108110eb
	VPERMR V1, V2, V3, V4           // 108110fb
	VPERMXOR V1, V2, V3, V4         // 108110ed
	VBPERMQ V1, V2, V3              // 1061154c
	VBPERMD V1, V2, V3              // 106115cc
	VSEL V1, V2, V3, V4             // 108110ea
	VSPLTB $1, V1, V2               // 10410a0c

	STXVH8X:        "stxvh8x",
	STXVL:          "stxvl",
	STXVLL:         "stxvll",
	STXVX:          "stxvx",
	VABSDUB:        "vabsdub",
	VABSDUH:        "vabsduh",
	VABSDUW:        "vabsduw",
	VBPERMD:        "vbpermd",
	VCLZLSBB:       "vclzlsbb",
	VCMPNEB:        "vcmpneb",
	VCMPNEBCC:      "vcmpneb.",
	VCMPNEH:        "vcmpneh",
	VCMPNEHCC:      "vcmpneh.",
	VCMPNEW:        "vcmpnew",

			opset(AVCMPNEBCC, r0)
			opset(AVCMPNEH, r0)
			opset(AVCMPNEHCC, r0)
			opset(AVCMPNEW, r0)
			opset(AVCMPNEWCC, r0)

		case AVPERM: /* vperm */
			opset(AVPERMXOR, r0)
			opset(AVPERMR, r0)

		case AVBPERMQ: /* vbpermq, vbpermd */
			opset(AVBPERMD, r0)

		case AVSEL: /* vsel */
			opset(AVSEL, r0)

		case AVSPLTB: /* vspltb, vsplth, vspltw */
			opset(AVSPLTH, r0)

	"VPDPWSSDS",
	"VPERM2F128",
	"VPERM2I128",
	"VPERMB",
	"VPERMD",
	"VPERMI2B",
	"VPERMI2D",
	"VPERMI2PD",
	"VPERMI2PS",
	"VPERMI2Q",
	"VPERMI2W",
	"VPERMILPD",
	"VPERMILPS",
	"VPERMPD",
	"VPERMPS",
	"VPERMQ",
	"VPERMT2B",
	"VPERMT2D",
	"VPERMT2PD",
	"VPERMT2PS",
	"VPERMT2Q",
	"VPERMT2W",
	"VPERMW",
	"VPEXPANDB",
	"VPEXPANDD",
	"VPEXPANDQ",

	ADD    $0x40, INP, INP
#ifdef GOARCH_ppc64le
	VPERM  IN0, IN0, LEMASK, IN0
	VPERM  IN1, IN1, LEMASK, IN1
	VPERM  IN2, IN2, LEMASK, IN2
	VPERM  IN3, IN3, LEMASK, IN3
#endif

	VXOR IN0, XL, XH

	VPMSUMD IN1, H3L, XL1
	VPMSUMD IN1, H3, XM1
	VPMSUMD IN1, H3H, XH1

	VPERM   H2, H, HIPERM, H21L
	VPERM   IN2, IN3, LOPERM, T0
	VPERM   H2, H, LOPERM, H21H
	VPERM   IN2, IN3, HIPERM, T1

#define NEEDS_ESPERM
#define P8_LXVB16X(RA,RB,VT) \
	LXVD2X	(RA+RB), VT \
	VPERM	VT, VT, ESPERM, VT

#define P8_STXVB16X(VS,RA,RB) \
	VPERM	VS, VS, ESPERM, TMP2 \
	STXVD2X	TMP2, (RA+RB)

#define XXBRD_ON_LE(VA,VT) \
	VPERM	VA, VA, ESPERM, VT

#  endif // defined(GOPPC64_power9)
#else
#define P8_LXVB16X(RA,RB,VT)  LXVD2X	(RA+RB), VT

#endif

// R3 = addr of string
// R4 = len of string
// R5 = byte to find
// On exit:
// R3 = return value
TEXT indexbytebody<>(SB),NOSPLIT|NOFRAME,$0-0
	CMPU	R4,$32

#ifndef GOPPC64_power9
	// Load VBPERMQ constant to reduce compare into an ordered bit mask.
	MOVD	$indexbytevbperm<>+00(SB),R16
	LXVD2X	(R16),V0	// Set up swap string
#endif

	MTVRD	R5,V1
	VSPLTB	$7,V1,V1	// Replicate byte across V1

	VADDUDM	V16, V6, V6
	VADDUDM	V17, V7, V7

	CMPU	INP, END
	BLT	loop

#ifdef GOARCH_ppc64le
	VPERM	V0, V1, KI, V0
	VPERM	V2, V3, KI, V2
	VPERM	V4, V5, KI, V4
	VPERM	V6, V7, KI, V6
#else
	VPERM	V1, V0, KI, V0
	VPERM	V3, V2, KI, V2
	VPERM	V5, V4, KI, V4
	VPERM	V7, V6, KI, V6
#endif
	STXVD2X	VS32, (CTX+R_x000)	// v0 = vs32
	STXVD2X	VS34, (CTX+R_x010)	// v2 = vs34

	SHA256MSG2 X2, X11       // 440f38cdda
	SHA256MSG2 X11, X11      // 450f38cddb
	// Test VPERMQ with both uint8 and int8 immediate args
	VPERMQ $-40, Y8, Y8 // c443fd00c0d8
	VPERMQ $216, Y8, Y8 // c443fd00c0d8
	// Test that VPERMPD that shares ytab list with VPERMQ continues to work too.
	VPERMPD $-40, Y7, Y7 // c4e3fd01ffd8
	VPERMPD $216, Y7, Y7 // c4e3fd01ffd8
	// Check that LEAL is permitted to use overflowing offset.