MOVD $32, R17
	MOVD $48, R18
	MOVD $56, R21
	MOVD $64, R19
	MOVD $80, R20
	// cond is R1 + 24 (cond offset) + 32
	LXVDSX (R1)(R21), SEL
	VSPLTISB $0, ZER
	// SEL controls whether to store a or b
	VCMPEQUD SEL, ZER, SEL

	LXVD2X (P1ptr+R0), X1H
	LXVD2X (P1ptr+R16), X1L
	LXVD2X (P1ptr+R17), Y1H
	LXVD2X (P1ptr+R18), Y1L
	LXVD2X (P1ptr+R19), Z1H
	LXVD2X (P1ptr+R20), Z1L

	LXVD2X (P2ptr+R0), X2H

	VPCMPEQQ Z20, Z1, K1, K4                           // 62b2f54929e4
	VPCMPEQQ Z9, Z1, K1, K4                            // 62d2f54929e1
	VPCMPEQQ (BX), Z1, K1, K4                          // 62f2f5492923
	VPCMPEQQ -17(BP)(SI*1), Z1, K1, K4                 // 62f2f54929a435efffffff
	VPCMPEQQ Z20, Z9, K1, K4                           // 62b2b54929e4
	VPCMPEQQ Z9, Z9, K1, K4                            // 62d2b54929e1

	PAVGW:           "PAVGW",
	PBLENDVB:        "PBLENDVB",
	PBLENDW:         "PBLENDW",
	PCLMULQDQ:       "PCLMULQDQ",
	PCMPEQB:         "PCMPEQB",
	PCMPEQD:         "PCMPEQD",
	PCMPEQQ:         "PCMPEQQ",
	PCMPEQW:         "PCMPEQW",
	PCMPESTRI:       "PCMPESTRI",
	PCMPESTRM:       "PCMPESTRM",
	PCMPGTB:         "PCMPGTB",
	PCMPGTD:         "PCMPGTD",
	PCMPGTQ:         "PCMPGTQ",