RET

cmp4:	// Length 4 - 7
	CMPU	R4,$4
	BLT	cmp2
	MOVD	$-4,R11
	ADD	$-4,R4,R4

	_LWBEX	(R0)(R3),R10
	_LWBEX	(R11)(R9),R11
	CMPB	R10,R5,R10
	CMPB	R11,R5,R11
	CNTLZW	R10,R10
	CNTLZW	R11,R11
	CMPU	R10,$32
	CMPU	R11,$32,CR1
	SRD	$3,R10,R3
	SRD	$3,R11,R11
	BNE	found

	ADD	R4,R11,R4
	MOVD	$-1,R3
	ISEL	CR1EQ,R3,R4,R3
	RET

cmp2:	// Length 2 - 3

#else
tail:	// Count the last 0 - 31 bytes.
	CMP	R4, $16
	BLT	tail_8
	MOVD	(R3), R12
	MOVD	8(R3), R14
	CMPB	R12, R5, R12
	CMPB	R14, R5, R14
	POPCNTD	R12, R12
	POPCNTD	R14, R14
	ADD	R12, R18, R18
	ADD	R14, R18, R18
	ADD	$16, R3, R3
	ADD	$-16, R4, R4

tail_8:	// Count the remaining 0 - 15 bytes.
	CMP	R4, $8
	BLT	tail_4
	MOVD	(R3), R12
	CMPB	R12, R5, R12
	POPCNTD	R12, R12
	ADD	R12, R18, R18

//   b in DI
//   count in BX
// Output:
//   result in AX
TEXT memeqbody<>(SB),NOSPLIT,$0-0
	CMPQ	BX, $8
	JB	small
	CMPQ	BX, $64
	JB	bigloop
#ifndef hasAVX2
	CMPB	internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
	JE	hugeloop_avx2

	// 64 bytes at a time using xmm registers
	PCALIGN $16
hugeloop:
	CMPQ	BX, $64
	JB	bigloop
	MOVOU	(SI), X0
	MOVOU	(DI), X1
	MOVOU	16(SI), X2
	MOVOU	16(DI), X3

	MOVL	-4(DI)(BX*1), DX
	CMPL	CX, DX
	SETEQ	(AX)
	RET

small:
	CMPL	BX, $0
	JEQ	equal

	LEAL	0(BX*8), CX
	NEGL	CX

	MOVL	SI, DX
	CMPB	DX, $0xfc
	JA	si_high

	// load at SI won't cross a page boundary.
	MOVL	(SI), SI
	JMP	si_finish
si_high:
	// address ends in 111111xx. Load up to bytes we want, move to correct position.
	MOVL	-4(SI)(BX*1), SI
	SHRL	CX, SI
si_finish:

#include "textflag.h"

TEXT ·Count(SB),NOSPLIT,$0-40
#ifndef hasPOPCNT
	CMPB	internal∕cpu·X86+const_offsetX86HasPOPCNT(SB), $1
	JEQ	2(PC)
	JMP	·countGeneric(SB)
#endif
	MOVQ	b_base+0(FP), SI
	MOVQ	b_len+8(FP), BX
	MOVB	c+24(FP), AL
	LEAQ	ret+32(FP), R8
	JMP	countbody<>(SB)

TEXT ·CountString(SB),NOSPLIT,$0-32
#ifndef hasPOPCNT
	CMPB	internal∕cpu·X86+const_offsetX86HasPOPCNT(SB), $1
	JEQ	2(PC)

	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:

TEXT cmpbody<>(SB),NOSPLIT,$0-0
	CMPQ	SI, DI
	JEQ	allsame
	CMPQ	BX, DX
	MOVQ	DX, R8
	CMOVQLT	BX, R8 // R8 = min(alen, blen) = # of bytes to compare
	CMPQ	R8, $8
	JB	small

	CMPQ	R8, $63
	JBE	loop
#ifndef hasAVX2
	CMPB	internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
	JEQ     big_loop_avx2
	JMP	big_loop
#else
	JMP	big_loop_avx2
#endif
loop:
	CMPQ	R8, $16
	JBE	_0through16
	MOVOU	(SI), X0

var x32 [2]uint32
var x64 [2]uint64

func compMem1() int {
	// amd64:`CMPB\tcommand-line-arguments.x\+1\(SB\), [$]0`
	if x[1] {
		return 1
	}
	// amd64:`CMPB\tcommand-line-arguments.x8\+1\(SB\), [$]7`
	if x8[1] == 7 {
		return 1
	}
	// amd64:`CMPW\tcommand-line-arguments.x16\+2\(SB\), [$]7`
	if x16[1] == 7 {
		return 1
	}
	// amd64:`CMPL\tcommand-line-arguments.x32\+4\(SB\), [$]7`
	if x32[1] == 7 {
		return 1
	}

	TESTQ	BX, BX
	JEQ	_0
	CMPQ	BX, $2
	JBE	_1or2
	CMPQ	BX, $4
	JBE	_3or4
	CMPQ	BX, $8
	JB	_5through7
	JE	_8
	CMPQ	BX, $16
	JBE	_9through16
	CMPQ	BX, $32
	JBE	_17through32
	CMPQ	BX, $64
	JBE	_33through64
	CMPQ	BX, $128
	JBE	_65through128
	CMPQ	BX, $256
	JBE	_129through256

	CMPB	internal∕cpu·X86+const_offsetX86HasERMS(SB), $1 // enhanced REP MOVSB/STOSB

(CMP(Q|L|W|B)loadidx1 {sym} [off] ptr idx x mem) => (CMP(Q|L|W|B) (MOV(Q|L|W|B)loadidx1 {sym} [off] ptr idx mem) x)
(CMPQloadidx8 {sym} [off] ptr idx x mem) => (CMPQ (MOVQloadidx8 {sym} [off] ptr idx mem) x)
(CMPLloadidx4 {sym} [off] ptr idx x mem) => (CMPL (MOVLloadidx4 {sym} [off] ptr idx mem) x)
(CMPWloadidx2 {sym} [off] ptr idx x mem) => (CMPW (MOVWloadidx2 {sym} [off] ptr idx mem) x)