VSHASIGMAW	$15, e, $1, S1; \
	VADDUWM		xi, h, h; \
	VSHASIGMAW	$0, a, $1, S0; \
	VSHASIGMAW	$15, xj_14, $0, s1; \
	VADDUWM		FUNC, h, h; \
	VXOR		b, a, FUNC; \
	VADDUWM		xj_9, xj, xj; \
	VADDUWM		S1, h, h; \
	VSEL		b, c, FUNC, FUNC; \
	VADDUWM		KI, g, g; \
	VADDUWM		h, d, d; \
	VADDUWM		FUNC, S0, S0; \
	VADDUWM		s0, xj, xj; \
	LVX		(TBL)(idx), KI; \
	VADDUWM		S0, h, h; \
	VADDUWM		s1, xj, xj

	VSLDOI	$12, ZERO, TMP, TMP // vsldoi 6,0,6,12
	VXOR	IN0, TMP, IN0       // vxor 1,1,6
	VSLDOI	$12, ZERO, TMP, TMP // vsldoi 6,0,6,12
	VXOR	IN0, TMP, IN0       // vxor 1,1,6
	VADDUWM	RCON, RCON, RCON    // vadduwm 4,4,4
	VXOR	IN0, KEY, IN0       // vxor 1,1,3
	BDNZ	loop128

	LVX	(PTR)(R0), RCON // lvx 4,0,6     Last two round keys

	// Key schedule (Round 9)

	VCMPEQUB V0, V1, V0
	VPOPCNTD V2, V2		// A match is 0xFF or 0. Count the bits into doubleword buckets.
	VPOPCNTD V0, V0
	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