MOVL	from+4(FP), SI
	MOVL	n+8(FP), BX

	// REP instructions have a high startup cost, so we handle small sizes
	// with some straightline code. The REP MOVSL instruction is really fast
	// 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

/*

TEXT runtime·setldt(SB),NOSPLIT,$32
	MOVL	base+4(FP), BX
	// see comment in sys_linux_386.s; freebsd is similar
	ADDL	$0x4, BX

	// set up data_desc
	LEAL	16(SP), AX	// struct data_desc
	MOVL	$0, 0(AX)
	MOVL	$0, 4(AX)

	MOVW	BX, 2(AX)
	SHRL	$16, BX
	MOVB	BX, 4(AX)
	SHRL	$8, BX
	MOVB	BX, 7(AX)

	MOVW	$0xffff, 0(AX)

	RET
overflow: // return +Inf
	MOVQ    $PosInf, BX
notNegInf: // NaN or +Inf, return x
	MOVQ    BX, ret+8(FP)
	RET
denormal:
	CMPL    BX, $-52
	JL      underflow
	ADDL    $0x3FE, BX // add bias - 1
	SHLQ    $52, BX
	MOVQ    BX, X1
	MULSD   X1, X0
	MOVQ    $1, BX
	JMP     lastStep

avxfma:
	MOVSD   $LN2U, X2
	VFNMADD231SD X2, X1, X0
	MOVSD   $LN2L, X2
	VFNMADD231SD X2, X1, X0

	PUSHL	AX
	PUSHL	BX
	get_tls(BX)
	MOVL	g(BX), BX
	LEAL	arg+0(FP), AX
	MOVL	AX, (g_sched+gobuf_sp)(BX)
	MOVL	$runtime·systemstack_switch(SB), AX
	MOVL	AX, (g_sched+gobuf_pc)(BX)
	MOVL	$0, (g_sched+gobuf_ret)(BX)
	// Assert ctxt is zero. See func save.
	MOVL	(g_sched+gobuf_ctxt)(BX), AX
	TESTL	AX, AX
	JZ	2(PC)
	CALL	runtime·abort(SB)
	POPL	BX
	POPL	AX
	RET

	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:
	MOVL	-4(SI)(BX*1), CX
	MOVL	-4(DI)(BX*1), DX
	CMPL	CX, DX

	MOVQ nr+0(FP), CX
	MOVQ key+8(FP), AX
	MOVQ enc+16(FP), BX
	MOVQ dec+24(FP), DX
	MOVUPS (AX), X0
	// enc
	MOVUPS X0, (BX)
	ADDQ $16, BX
	PXOR X4, X4 // _expand_key_* expect X4 to be zero
	CMPL CX, $12
	JE Lexp_enc192
	JB Lexp_enc128
Lexp_enc256:
	MOVUPS 16(AX), X2
	MOVUPS X2, (BX)
	ADDQ $16, BX
	AESKEYGENASSIST $0x01, X2, X1
	CALL _expand_key_256a<>(SB)

	ROUND2(BX, CX, DX, BP, AX, 34)
	ROUND2(AX, BX, CX, DX, BP, 35)
	ROUND2(BP, AX, BX, CX, DX, 36)
	ROUND2(DX, BP, AX, BX, CX, 37)
	ROUND2(CX, DX, BP, AX, BX, 38)
	ROUND2(BX, CX, DX, BP, AX, 39)

	ROUND3(AX, BX, CX, DX, BP, 40)
	ROUND3(BP, AX, BX, CX, DX, 41)
	ROUND3(DX, BP, AX, BX, CX, 42)
	ROUND3(CX, DX, BP, AX, BX, 43)
	ROUND3(BX, CX, DX, BP, AX, 44)
	ROUND3(AX, BX, CX, DX, BP, 45)

//   SI: data
//   BX: data len
//   AL: byte sought
//   R8: address to put result
// This function requires the POPCNT instruction.
TEXT countbody<>(SB),NOSPLIT,$0
	// Shuffle X0 around so that each byte contains
	// the character we're looking for.
	MOVD AX, X0
	PUNPCKLBW X0, X0
	PUNPCKLBW X0, X0
	PSHUFL $0, X0, X0

	CMPQ BX, $16
	JLT small

	MOVQ $0, R12 // Accumulator

	MOVQ SI, DI

	CMPQ BX, $64
	JAE avx2

	tstb()
	by = bx[0:10]
	tstb()
	by = bx[0:]
	tstb()
	by = bx[:10]
	tstb()
	by = bx[:]
	tstb()

	lb = 2
	hb = 10
	by = bx[lb:hb]
	tstb()
	by = bx[lb:10]
	tstb()
	by = bx[lb:]
	tstb()
	by = bx[2:hb]
	tstb()
	by = bx[2:10]
	tstb()
	by = bx[2:]
	tstb()

	lb = 0
	hb = 8
	by = bx[lb:hb]
	tstb()
	by = bx[lb:8]
	tstb()
	by = bx[0:hb]
	tstb()
	by = bx[0:8]
	tstb()

TEXT	·IndexByte(SB), NOSPLIT, $0-40
	MOVQ b_base+0(FP), SI
	MOVQ b_len+8(FP), BX
	MOVB c+24(FP), AL
	LEAQ ret+32(FP), R8
	JMP  indexbytebody<>(SB)

TEXT	·IndexByteString(SB), NOSPLIT, $0-32
	MOVQ s_base+0(FP), SI
	MOVQ s_len+8(FP), BX
	MOVB c+16(FP), AL
	LEAQ ret+24(FP), R8
	JMP  indexbytebody<>(SB)

// input:
//   SI: data
//   BX: data len
//   AL: byte sought
//   R8: address to put result