SHA512H	V7.D2, V6, i3 \
	VADD	i3.D2, i1.D2, i4.D2 \
	SHA512H2	i0.D2, i1, i3

// func blockAsm(dig *digest, p []byte)
TEXT ·blockAsm(SB),NOSPLIT,$0
	MOVD	dig+0(FP), R0
	MOVD	p_base+8(FP), R1
	MOVD	p_len+16(FP), R2
	MOVD	·_K+0(SB), R3

	// long enough to prefetch
	PRFM	(R3), PLDL3KEEP
	// load digest
	VLD1	(R0), [V8.D2, V9.D2, V10.D2, V11.D2]
loop:

#define SHA512ROUND1(index, a, b, c, d, e, f, g, h) \
	MSGSCHEDULE1(index); \
	SHA512ROUND(index, a, b, c, d, e, f, g, h)

// func block(dig *digest, p []byte)
TEXT ·block(SB),0,$128-32
	MOV	p_base+8(FP), X29
	MOV	p_len+16(FP), X30
	SRL	$7, X30
	SLL	$7, X30

	ADD	X29, X30, X28
	BEQ	X28, X29, end

	MOV	·_K(SB), X18		// const table
	ADD	$8, X2, X19		// message schedule