VADDUDM	V0, V4, V4	// Accumulate the popcounts. They are 8x the count.
	VADDUDM	V2, V5, V5	// The count will be fixed up afterwards.
	ADD	$32, R3
	BDNZ	cmploop

	VADDUDM	V4, V5, V5
	MFVSRD	V5, R18
	VSLDOI	$8, V5, V5, V5
	MFVSRD	V5, R21
	ADD	R21, R18, R18
	ANDCC	$31, R4, R4
	// Skip the tail processing if no bytes remaining.
	BEQ	tail_0

#ifdef GOPPC64_power10

	MFVSRD	VS36,R10

	CMPU	R16,R10
	BEQ	lower
	SETB_CR0_NE(R3)
	RET

	PCALIGN $16
lower:
	VSLDOI	$8,V3,V3,V3	// move lower doublewords of A and B into GPR for comparison
	MFVSRD	VS35,R16
	VSLDOI	$8,V4,V4,V4
	MFVSRD	VS36,R10

	CMPU	R16,R10
	SETB_CR0_NE(R3)
	RET

	PCALIGN $16
cmp8:	// 8 - 15B (0 - 15B if GOPPC64_power10)

	"VSLH",
	"VSLW",
	"VSL",
	"VSLO",
	"VSRB",
	"VSRH",
	"VSRW",
	"VSR",
	"VSRO",
	"VSLD",
	"VSRD",
	"VSA",
	"VSRAB",
	"VSRAH",
	"VSRAW",
	"VSRAD",
	"VSOI",
	"VSLDOI",
	"VCLZ",
	"VCLZB",
	"VCLZH",
	"VCLZW",
	"VCLZD",
	"VPOPCNT",
	"VPOPCNTB",
	"VPOPCNTH",
	"VPOPCNTW",
	"VPOPCNTD",
	"VCMPEQ",
	"VCMPEQUB",
	"VCMPEQUBCC",
	"VCMPEQUH",

	// V28
	LXVW4X (CONSTBASE)(R11), VS60

	// Load mask constants for VPERMXOR
	LXVW4X (MASK)(R0), V20
	LXVW4X (MASK)(R20), V21

	// splat slot from V19 -> V26
	VSPLTW $0, V19, V26

	VSLDOI $4, V19, V27, V19
	VSLDOI $12, V27, V19, V19

	VADDUWM V26, V28, V26

	MOVD $10, R14
	MOVD R14, CTR
	PCALIGN $16
loop_outer_vsx:
	// V0, V1, V2, V3
	LXVW4X (R0)(CONSTBASE), VS32