MOV	$8, X9
	BLT	X12, X9, f_loop4_check
f_loop8_unaligned:
	MOVB	0(X11), X14
	MOVB	1(X11), X15
	MOVB	2(X11), X16
	MOVB	3(X11), X17
	MOVB	4(X11), X18
	MOVB	5(X11), X19
	MOVB	6(X11), X20
	MOVB	7(X11), X21
	MOVB	X14, 0(X10)
	MOVB	X15, 1(X10)
	MOVB	X16, 2(X10)
	MOVB	X17, 3(X10)
	MOVB	X18, 4(X10)
	MOVB	X19, 5(X10)
	MOVB	X20, 6(X10)
	MOVB	X21, 7(X10)
	ADD	$8, X10
	ADD	$8, X11
	SUB	$8, X12

	"MODW",
	"MODWU",
	"MODD",
	"MODDU",
	"MULLW",
	"MULLD",
	"MULHD",
	"MULHDU",
	"MLGR",
	"SUB",
	"SUBC",
	"SUBV",
	"SUBE",
	"SUBW",
	"NEG",
	"NEGW",
	"MOVWBR",
	"MOVB",
	"MOVBZ",
	"MOVH",
	"MOVHBR",
	"MOVHZ",
	"MOVW",
	"MOVWZ",
	"MOVD",
	"MOVDBR",
	"MOVDEQ",
	"MOVDGE",
	"MOVDGT",
	"MOVDLE",
	"MOVDLT",
	"MOVDNE",
	"LOCR",
	"LOCGR",
	"FLOGR",

	RET

// void ·Or8(byte volatile*, byte);
TEXT ·Or8(SB), NOSPLIT, $0-9
	MOVD	ptr+0(FP), R3
	MOVBZ	val+8(FP), R4
	LWSYNC
again:
	LBAR	(R3), R6
	OR	R4, R6
	STBCCC	R6, (R3)
	BNE	again
	RET

// void ·And8(byte volatile*, byte);
TEXT ·And8(SB), NOSPLIT, $0-9
	MOVD	ptr+0(FP), R3
	MOVBZ	val+8(FP), R4
	LWSYNC
again:
	LBAR	(R3), R6
	AND	R4, R6
	STBCCC	R6, (R3)
	BNE	again
	RET

	CMP	R0, R2
	BNE	casfail

#ifndef GOARM_7
	MOVB	internal∕cpu·ARM+const_offsetARMHasV7Atomics(SB), R11
	CMP	$0, R11
	BEQ	2(PC)
#endif
	DMB	MB_ISHST

	STREX	R3, (R1), R0
	CMP	$0, R0
	BNE	casl
	MOVW	$1, R0

#ifndef GOARM_7
	CMP	$0, R11
	BEQ	2(PC)
#endif
	DMB	MB_ISH

	MOVB	R0, ret+12(FP)
	RET
casfail:
	MOVW	$0, R0
	MOVB	R0, ret+12(FP)
	RET

// stubs

	SW	X6, 4096(X5)
	SD	X6, 4096(X5)

	FLW	4096(X5), F6
	FLD	4096(X5), F6
	FSW	F6, 4096(X5)
	FSD	F6, 4096(X5)

	MOVB	4096(X5), X6
	MOVH	4096(X5), X6
	MOVW	4096(X5), X6
	MOV	4096(X5), X6
	MOVBU	4096(X5), X6
	MOVHU	4096(X5), X6
	MOVWU	4096(X5), X6

	MOVB	X6, 4096(X5)
	MOVH	X6, 4096(X5)
	MOVW	X6, 4096(X5)
	MOV	X6, 4096(X5)

	MOVF	4096(X5), F6
	MOVD	4096(X5), F6

	}

	// It must be followed by a MOVQ SP, BP.
	movbp := pushbp.Link
	if movbp == nil {
		ctxt.Diag("missing frame pointer instruction: MOVQ SP, BP")
		return
	}
	if !(movbp.As == AMOVQ && movbp.From.Type == obj.TYPE_REG && movbp.From.Reg == REG_SP &&
		movbp.To.Type == obj.TYPE_REG && movbp.To.Reg == REG_BP && movbp.From.Offset == 0) {
		ctxt.Diag("unexpected frame pointer instruction\n%v", movbp)
		return
	}

	SC	R3, (R1)	// *R1 = R3
	BEQ	R3, try_cas
	SYNC
	MOVB	R3, ret+12(FP)
	RET
cas_fail:
	SYNC
	MOVB	R0, ret+12(FP)
	RET

TEXT ·Store(SB),NOSPLIT,$0-8
	MOVW	ptr+0(FP), R1
	MOVW	val+4(FP), R2
	SYNC
	MOVW	R2, 0(R1)
	SYNC
	RET

TEXT ·Store8(SB),NOSPLIT,$0-5
	MOVW	ptr+0(FP), R1
	MOVB	val+4(FP), R2
	SYNC
	MOVB	R2, 0(R1)
	SYNC
	RET

TEXT ·Load(SB),NOSPLIT,$0-8

	RET

TEXT ·Store8(SB), NOSPLIT, $0-9
	MOVV	ptr+0(FP), R4
	MOVB	val+8(FP), R5
	DBAR
	MOVB	R5, 0(R4)
	DBAR
	RET

TEXT ·Store64(SB), NOSPLIT, $0-16
	MOVV	ptr+0(FP), R4
	MOVV	val+8(FP), R5
	DBAR
	MOVV	R5, 0(R4)
	DBAR
	RET

// void	Or8(byte volatile*, byte);
TEXT ·Or8(SB), NOSPLIT, $0-9
	MOVV	ptr+0(FP), R4
	MOVBU	val+8(FP), R5

#include "go_asm.h"
#include "textflag.h"

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

#include "go_asm.h"
#include "textflag.h"

TEXT ·Count(SB),NOSPLIT,$0-20
	MOVW	b_base+0(FP), R0
	MOVW	b_len+4(FP), R1
	MOVBU	c+12(FP), R2
	MOVW	$ret+16(FP), R7
	B	countbytebody<>(SB)

TEXT ·CountString(SB),NOSPLIT,$0-16
	MOVW	s_base+0(FP), R0
	MOVW	s_len+4(FP), R1
	MOVBU	c+8(FP), R2
	MOVW	$ret+12(FP), R7
	B	countbytebody<>(SB)

// Input:
// R0: data
// R1: data length
// R2: byte to find