PAND	X6, X4
	PAND	X4, X0
	PMOVMSKB X0, DX
	ADDL	$64, SI
	ADDL	$64, DI
	SUBL	$64, BX
	CMPL	DX, $0xffff
	JEQ	hugeloop
	MOVB	$0, (AX)
	RET

	// 4 bytes at a time using 32-bit register
bigloop:
	CMPL	BX, $4
	JBE	leftover
	MOVL	(SI), CX
	MOVL	(DI), DX
	ADDL	$4, SI
	ADDL	$4, DI
	SUBL	$4, BX
	CMPL	CX, DX
	JEQ	bigloop
	MOVB	$0, (AX)
	RET

	// remaining 0-4 bytes
leftover:

// See issue 43661.

TEXT ·a1(SB), 0, $0-0
	CMPL runtime·writeBarrier(SB), $0
	MOVL $0, R15
	RET
TEXT ·a2(SB), 0, $0-0
	CMPL runtime·writeBarrier(SB), $0
	MOVQ $0, R15
	RET
TEXT ·a3(SB), 0, $0-0
	CMPL runtime·writeBarrier(SB), $0
	XORL R15, R15
	RET
TEXT ·a4(SB), 0, $0-0
	CMPL runtime·writeBarrier(SB), $0
	XORQ R15, R15
	RET
TEXT ·a5(SB), 0, $0-0
	CMPL runtime·writeBarrier(SB), $0
	XORL R15, R15
	RET

	TESTL	BX, BX
	JEQ	move_0
	CMPL	BX, $2
	JBE	move_1or2
	CMPL	BX, $4
	JB	move_3
	JE	move_4
	CMPL	BX, $8
	JBE	move_5through8
	CMPL	BX, $16
	JBE	move_9through16
#ifdef GO386_softfloat
	JMP	nosse2
#endif
	CMPL	BX, $32
	JBE	move_17through32
	CMPL	BX, $64
	JBE	move_33through64
	CMPL	BX, $128
	JBE	move_65through128

nosse2:
/*

	TESTL	BX, BX
	JEQ	_0
	CMPL	BX, $2
	JBE	_1or2
	CMPL	BX, $4
	JB	_3
	JE	_4
	CMPL	BX, $8
	JBE	_5through8
	CMPL	BX, $16
	JBE	_9through16
#ifdef GO386_softfloat
	JMP	nosse2
#endif
	PXOR	X0, X0
	CMPL	BX, $32
	JBE	_17through32
	CMPL	BX, $64
	JBE	_33through64
	CMPL	BX, $128
	JBE	_65through128
	CMPL	BX, $256
	JBE	_129through256

loop:
	MOVOU	X0, 0(DI)

	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	-4(SI)(BP*1), BX
	MOVL	-4(DI)(BP*1), CX
	CMPL	BX, CX
	JEQ	allsame

diff4:

func si1c(c <-chan int64) {
	// amd64:"CMPQ\t.+, [$]256"
	// s390x:"CLGIJ\t[$]12, R[0-9]+, [$]255"
	for x := <-c; x >= 0 && x < 256; x = <-c {
	}
}

func si2c(c <-chan int32) {
	// amd64:"CMPL\t.+, [$]256"
	// s390x:"CLIJ\t[$]12, R[0-9]+, [$]255"
	for x := <-c; x >= 0 && x < 256; x = <-c {
	}
}

func si3c(c <-chan int16) {
	// amd64:"CMPW\t.+, [$]256"
	// s390x:"CLIJ\t[$]12, R[0-9]+, [$]255"

	// for large sizes. The cutover is approximately 1K.
tail:
	TESTL	BX, BX
	JEQ	move_0
	CMPL	BX, $2
	JBE	move_1or2
	CMPL	BX, $4
	JB	move_3
	JE	move_4
	CMPL	BX, $8
	JBE	move_5through8
	CMPL	BX, $16
	JBE	move_9through16

/*
 * check and set for backwards
 */
	CMPL	SI, DI
	JLS	back

/*
 * forward copy loop
 */
forward:
	MOVL	BX, CX
	SHRL	$2, CX
	ANDL	$3, BX

loop3:
	MOVW (DI), SI
	CMPW SI,R8
	JZ   partial_success3
	ADDQ $1,DI
	CMPQ DI,DX
	JB loop3
	JMP fail
partial_success3:
	MOVW 1(DI), SI
	CMPW SI,BX
	JZ success
	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

// func archHypot(p, q float64) float64
TEXT ·archHypot(SB),NOSPLIT,$0
// test bits for not-finite
	MOVL    p_hi+4(FP), AX   // high word p
	ANDL    $0x7ff00000, AX
	CMPL    AX, $0x7ff00000
	JEQ     not_finite
	MOVL    q_hi+12(FP), AX   // high word q
	ANDL    $0x7ff00000, AX
	CMPL    AX, $0x7ff00000
	JEQ     not_finite
	FMOVD   p+0(FP), F0  // F0=p
	FABS                 // F0=|p|
	FMOVD   q+8(FP), F0  // F0=q, F1=|p|

TEXT runtime·memclrNoHeapPointers(SB), NOSPLIT, $0-8
	MOVL	ptr+0(FP), DI
	MOVL	n+4(FP), BX
	XORL	AX, AX

tail:
	TESTL	BX, BX
	JEQ	_0
	CMPL	BX, $2
	JBE	_1or2
	CMPL	BX, $4
	JB	_3
	JE	_4
	CMPL	BX, $8
	JBE	_5through8
	CMPL	BX, $16
	JBE	_9through16
	MOVL	BX, CX
	SHRL	$2, CX
	REP
	STOSL
	ANDL	$3, BX
	JNE	tail
	RET

_1or2:
	MOVB	AX, (DI)
	MOVB	AX, -1(DI)(BX*1)