"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",

	VPERMD (DX), Z0, K1, Z7                            // 62f27d49363a
	VPERMD Z20, Z6, K1, Z7                             // 62b24d4936fc
	VPERMD Z28, Z6, K1, Z7                             // 62924d4936fc
	VPERMD 99(R15)(R15*1), Z6, K1, Z7                  // 62924d4936bc3f63000000
	VPERMD (DX), Z6, K1, Z7                            // 62f24d49363a
	VPERMD Z20, Z0, K1, Z9                             // 62327d4936cc

	VPERMB X19, X2, K1, X23                            // 62a26d098dfb
	VPERMB X0, X2, K1, X23                             // 62e26d098df8
	VPERMB 7(SI)(DI*4), X2, K1, X23                    // 62e26d098dbcbe07000000
	VPERMB -7(DI)(R8*2), X2, K1, X23                   // 62a26d098dbc47f9ffffff
	VPERMB X26, X9, K1, X23                            // 628235098dfa
	VPERMB X19, X9, K1, X23                            // 62a235098dfb

	VLM     0(R3), V1, V4    // 64-bytes into V1..V4

	// Reflect the data if the CRC operation is in the bit-reflected domain
	VPERM   V1, V1, CONST_PERM_LE2BE, V1
	VPERM   V2, V2, CONST_PERM_LE2BE, V2
	VPERM   V3, V3, CONST_PERM_LE2BE, V3
	VPERM   V4, V4, CONST_PERM_LE2BE, V4

	VX      V0, V1, V1     // V1 ^= CRC
	ADD     $64, R3        // BUF = BUF + 64
	ADD     $(-64), R4

	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