// We may add checks inside blockAVX2, but this will
		// just turn it into a copy of blockAMD64,
		// so call it directly, instead.
		safeLen := len(p) - 128
		if safeLen%128 != 0 {
			safeLen -= 64
		}
		blockAVX2(dig, p[:safeLen])
		blockAMD64(dig, p[safeLen:])
	} else {
		blockAMD64(dig, p)
	}

package sha512

import "internal/cpu"

//go:noescape
func blockAVX2(dig *digest, p []byte)

//go:noescape
func blockAMD64(dig *digest, p []byte)

var useAVX2 = cpu.X86.HasAVX2 && cpu.X86.HasBMI1 && cpu.X86.HasBMI2

func block(dig *digest, p []byte) {
	if useAVX2 {
		blockAVX2(dig, p)
	} else {
		blockAMD64(dig, p)
	}

	SHUFFLE(index); \
	FUNC3(a, b, c, d, e); \
	MIX(a, b, c, d, e, 0x8F1BBCDC)

#define ROUND4(a, b, c, d, e, index) \
	SHUFFLE(index); \
	FUNC4(a, b, c, d, e); \
	MIX(a, b, c, d, e, 0xCA62C1D6)

TEXT ·blockAMD64(SB),NOSPLIT,$64-32
	MOVQ	dig+0(FP),	BP
	MOVQ	p_base+8(FP),	SI
	MOVQ	p_len+16(FP),	DX
	SHRQ	$6,		DX
	SHLQ	$6,		DX

	LEAQ	(SI)(DX*1),	DI
	MOVL	(0*4)(BP),	AX
	MOVL	(1*4)(BP),	BX

	SHA512ROUND(index, const, a, b, c, d, e, f, g, h)

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

TEXT ·blockAMD64(SB),0,$648-32
	MOVQ	p_base+8(FP), SI
	MOVQ	p_len+16(FP), DX
	SHRQ	$7, DX
	SHLQ	$7, DX

	LEAQ	(SI)(DX*1), DI
	MOVQ	DI, 640(SP)
	CMPQ	SI, DI
	JEQ	end

	MOVQ	dig+0(FP), BP