MOVB  (SI)(DX*1), DI
	MOVB  (CX)(DX*1), AX
	XORB  AX, DI
	MOVB  DI, (BX)(DX*1)
	TESTQ $7, DX           // AND 7 & len, if not zero jump to loop_1b.
	JNZ   loop_1b
	CMPQ  DX, $0           // if len is 0, ret.
	JE    ret
	TESTQ $15, DX          // AND 15 & len, if zero jump to aligned.
	JZ    aligned

not_aligned:
	TESTQ $7, DX           // AND $7 & len, if not zero jump to loop_1b.

	MOVL x+12(FP), SI
	MOVL y+24(FP), CX
	MOVL z_len+4(FP), BP
	MOVL $0, BX		// i = 0
	MOVL $0, DX		// c = 0
	JMP E1

L1:	MOVL (SI)(BX*4), AX
	ADDL DX, DX		// restore CF
	ADCL (CX)(BX*4), AX
	SBBL DX, DX		// save CF
	MOVL AX, (DI)(BX*4)
	ADDL $1, BX		// i++

E1:	CMPL BX, BP		// i < n
	JL L1

	NEGL DX
	MOVL DX, c+36(FP)
	RET


// func subVV(z, x, y []Word) (c Word)

	ROUND1(CX,DX,AX,BX, 3,0x242070db,17);
	ROUND1(BX,CX,DX,AX, 4,0xc1bdceee,22);
	ROUND1(AX,BX,CX,DX, 5,0xf57c0faf, 7);
	ROUND1(DX,AX,BX,CX, 6,0x4787c62a,12);
	ROUND1(CX,DX,AX,BX, 7,0xa8304613,17);
	ROUND1(BX,CX,DX,AX, 8,0xfd469501,22);
	ROUND1(AX,BX,CX,DX, 9,0x698098d8, 7);
	ROUND1(DX,AX,BX,CX,10,0x8b44f7af,12);
	ROUND1(CX,DX,AX,BX,11,0xffff5bb1,17);
	ROUND1(BX,CX,DX,AX,12,0x895cd7be,22);

	ADDL	$16, DI
	SUBL	$16, BP
	JMP	largeloop

diff16:
	BSFL	BX, BX	// index of first byte that differs
	XORL	DX, DX
	MOVB	(SI)(BX*1), CX
	CMPB	CX, (DI)(BX*1)
	SETHI	DX
	LEAL	-1(DX*2), DX	// convert 1/0 to +1/-1
	MOVL	DX, (AX)
	RET

mediumloop:
	CMPL	BP, $4
	JBE	_0through4
	MOVL	(SI), BX
	MOVL	(DI), CX
	CMPL	BX, CX
	JNE	diff4
	ADDL	$4, SI
	ADDL	$4, DI
	SUBL	$4, BP
	JMP	mediumloop

_0through4:

	MOVL	c, R9; \
	ADDL	b, a

	ROUND1(AX,BX,CX,DX, 1,0xd76aa478, 7);
	ROUND1(DX,AX,BX,CX, 2,0xe8c7b756,12);
	ROUND1(CX,DX,AX,BX, 3,0x242070db,17);
	ROUND1(BX,CX,DX,AX, 4,0xc1bdceee,22);
	ROUND1(AX,BX,CX,DX, 5,0xf57c0faf, 7);
	ROUND1(DX,AX,BX,CX, 6,0x4787c62a,12);
	ROUND1(CX,DX,AX,BX, 7,0xa8304613,17);
	ROUND1(BX,CX,DX,AX, 8,0xfd469501,22);
	ROUND1(AX,BX,CX,DX, 9,0x698098d8, 7);
	ROUND1(DX,AX,BX,CX,10,0x8b44f7af,12);

	ROUND1(BP, AX, BX, CX, DX, 1)
	ROUND1(DX, BP, AX, BX, CX, 2)
	ROUND1(CX, DX, BP, AX, BX, 3)
	ROUND1(BX, CX, DX, BP, AX, 4)
	ROUND1(AX, BX, CX, DX, BP, 5)
	ROUND1(BP, AX, BX, CX, DX, 6)
	ROUND1(DX, BP, AX, BX, CX, 7)
	ROUND1(CX, DX, BP, AX, BX, 8)
	ROUND1(BX, CX, DX, BP, AX, 9)
	ROUND1(AX, BX, CX, DX, BP, 10)
	ROUND1(BP, AX, BX, CX, DX, 11)
	ROUND1(DX, BP, AX, BX, CX, 12)
	ROUND1(CX, DX, BP, AX, BX, 13)

	ADDQ $1,DI
	CMPQ DI,DX
	JB loop3
	JMP fail
_4_or_more:
	CMPQ AX, $4
	JA   _5_or_more
	MOVL (R8), R8
	LEAQ -3(DI)(DX*1), DX
loop4:
	MOVL (DI), SI
	CMPL SI,R8
	JZ   success
	ADDQ $1,DI
	CMPQ DI,DX
	JB loop4
	JMP fail
_5_or_more:
	CMPQ AX, $7
	JA   _8_or_more
	LEAQ 1(DI)(DX*1), DX
	SUBQ AX, DX
	MOVL -4(R8)(AX*1), BX
	MOVL (R8), R8
loop5to7:
	MOVL (DI), SI
	CMPL SI,R8

	//	func f(x uint64) (uint64, uint64) { return x/1000000000, x%1000000000 }
	// adapted to reduce duplication
	MOVQ	AX, CX
	MOVQ	$1360296554856532783, AX
	MULQ	CX
	ADDQ	CX, DX
	RCRQ	$1, DX
	SHRQ	$29, DX
	MOVQ	DX, sec+0(FP)
	IMULQ	$1000000000, DX
	SUBQ	DX, CX
	MOVL	CX, nsec+8(FP)
	RET

TEXT runtime·notify(SB),NOSPLIT,$0
	MOVQ	$28, BP
	SYSCALL
	MOVL	AX, ret+8(FP)
	RET

	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (0*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (1*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (1*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (2*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (2*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (3*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (3*8)(R10)(BX*8)
	MOVQ DX, CX

	PMOVMSKB X1, DX	// Move result bits to integer register.
	MOVL	BX, CX
	SHLL	CX, DX
	SHRL	$16, DX	// Shift desired bits down to bottom of register.
	BSFL	DX, DX	// Find first set bit.
	JZ	failure	// No set bit, failure.
	MOVQ	DX, (R8)
	RET

avx2:
#ifndef hasAVX2
	CMPB   internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
	JNE sse
#endif
	MOVD AX, X0
	LEAQ -32(SI)(BX*1), R11