MOVD	$AT_FDCWD, R0
	MOVD	name+0(FP), R1
	MOVW	mode+8(FP), R2
	MOVW	perm+12(FP), R3
	MOVD	$SYS_openat, R8
	SVC
	CMN	$4095, R0
	BCC	done
	MOVW	$-1, R0
done:
	MOVW	R0, ret+16(FP)
	RET

TEXT runtime·closefd(SB),NOSPLIT|NOFRAME,$0-12
	MOVW	fd+0(FP), R0
	MOVD	$SYS_close, R8
	SVC
	CMN	$4095, R0
	BCC	done
	MOVW	$-1, R0
done:
	MOVW	R0, ret+8(FP)
	RET

	MOVV	HI, R1		// 00000810
	MOVV	R1, LO		// 00200013
	MOVV	R1, HI		// 00200011


//	LMOVW rreg ',' rreg
//	{
//		outcode(int($1), &$2, 0, &$4);
//	}
	MOVW	R1, R2		// 00011004
	MOVW	LO, R1		// 00000812
	MOVW	HI, R1		// 00000810
	MOVW	R1, LO		// 00200013
	MOVW	R1, HI		// 00200011
	MOVWU	R14, R27	// 000ed83c001bd83e

//	LMOVH rreg ',' rreg
//	{
//		outcode(int($1), &$2, 0, &$4);
//	}

TEXT armCas64<>(SB),NOSPLIT,$0-21
	// addr is already in R1
	MOVW	old_lo+4(FP), R2
	MOVW	old_hi+8(FP), R3
	MOVW	new_lo+12(FP), R4
	MOVW	new_hi+16(FP), R5
cas64loop:
	LDREXD	(R1), R6	// loads R6 and R7
	CMP	R2, R6
	BNE	cas64fail
	CMP	R3, R7
	BNE	cas64fail

	DMB	MB_ISHST

	STREXD	R4, (R1), R0	// stores R4 and R5
	CMP	$0, R0
	BNE	cas64loop
	MOVW	$1, R0

	DMB	MB_ISH

	MOVBU	R0, swapped+20(FP)

TEXT runtime·memequal(SB),NOSPLIT|NOFRAME,$0-13
	MOVW	a+0(FP), R0
	MOVW	b+4(FP), R2
	CMP	R0, R2
	B.EQ	eq
	MOVW	size+8(FP), R1
	CMP	$0, R1
	B.EQ	eq		// short path to handle 0-byte case
	MOVW	$ret+12(FP), R7
	B	memeqbody<>(SB)
eq:
	MOVW	$1, R0
	MOVB	R0, ret+12(FP)
	RET

// memequal_varlen(a, b unsafe.Pointer) bool
TEXT runtime·memequal_varlen(SB),NOSPLIT|NOFRAME,$0-9
	MOVW	a+0(FP), R0
	MOVW	b+4(FP), R2
	CMP	R0, R2

TEXT memory_barrier<>(SB),NOSPLIT|NOFRAME,$0
	MOVW	$0xffff0fa0, R15 // R15 is hardware PC.

TEXT	·Load(SB),NOSPLIT,$0-8
	MOVW	addr+0(FP), R0
	MOVW	(R0), R1

	MOVB	runtime·goarm(SB), R11
	CMP	$7, R11
	BGE	native_barrier
	BL	memory_barrier<>(SB)
	B	end
native_barrier:
	DMB	MB_ISH
end:
	MOVW	R1, ret+4(FP)
	RET

TEXT	·Store(SB),NOSPLIT,$0-8
	MOVW	addr+0(FP), R1
	MOVW	v+4(FP), R2

	MOVB	runtime·goarm(SB), R8

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

TEXT ·IndexByte(SB),NOSPLIT,$0-20
	MOVW	b_base+0(FP), R0
	MOVW	b_len+4(FP), R1
	MOVBU	c+12(FP), R2	// byte to find
	MOVW	$ret+16(FP), R5
	B	indexbytebody<>(SB)

TEXT ·IndexByteString(SB),NOSPLIT,$0-16
	MOVW	s_base+0(FP), R0
	MOVW	s_len+4(FP), R1
	MOVBU	c+8(FP), R2	// byte to find
	MOVW	$ret+12(FP), R5
	B	indexbytebody<>(SB)

// input:
//  R0: data
//  R1: data length

	RET

TEXT ·IndexByteString(SB),NOSPLIT,$0-16
	MOVW	s_base+0(FP), R1
	MOVW	s_len+4(FP), R2
	MOVBU	c+8(FP), R3	// byte to find
	ADDU	$1, R1, R4	// store base+1 for later
	ADDU	R1, R2	// end

loop:
	BEQ	R1, R2, notfound
	MOVBU	(R1), R5
	ADDU	$1, R1
	BNE	R3, R5, loop

	SUBU	R4, R1	// remove (base+1)
	MOVW	R1, ret+12(FP)
	RET

notfound:
	MOVW	$-1, R1
	MOVW	R1, ret+12(FP)

	SUBU	R5, R4, R5	// end pointer for 4-byte chunks

large:
	BEQ	R1, R6, words
	MOVW	R0, 0(R1)
	MOVW	R0, 4(R1)
	MOVW	R0, 8(R1)
	MOVW	R0, 12(R1)
	MOVW	R0, 16(R1)
	MOVW	R0, 20(R1)
	MOVW	R0, 24(R1)
	MOVW	R0, 28(R1)
	ADDU	$32, R1
	JMP	large

words:
	BEQ	R1, R5, tail
	MOVW	R0, 0(R1)
	ADDU	$4, R1
	JMP	words

tail:
	BEQ	R1, R4, ret
	MOVWLO	R0, -1(R4)

ret:
	RET

TEXT runtime·memequal(SB),NOSPLIT,$0-13
	MOVW	a+0(FP), R1
	MOVW	b+4(FP), R2
	BEQ	R1, R2, eq
	MOVW	size+8(FP), R3
	ADDU	R1, R3, R4
loop:
	BNE	R1, R4, test
	MOVW	$1, R1
	MOVB	R1, ret+12(FP)
	RET
test:
	MOVBU	(R1), R6
	ADDU	$1, R1
	MOVBU	(R2), R7
	ADDU	$1, R2
	BEQ	R6, R7, loop

	MOVB	R0, ret+12(FP)
	RET
eq:
	MOVW	$1, R1
	MOVB	R1, ret+12(FP)
	RET

	LDPW	(1*8)(R0), (R6, R7)

loop:
	MOVW	R4, R12
	MOVW	R5, R13
	MOVW	R6, R14
	MOVW	R7, R15

	MOVW	(0*4)(R1), R8
	MOVW	R7, R9

#define ROUND1(a, b, c, d, index, const, shift) \
	ADDW	$const, a; \
	ADDW	R8, a; \
	MOVW	(index*4)(R1), R8; \
	EORW	c, R9; \
	ANDW	b, R9; \
	EORW	d, R9; \
	ADDW	R9, a; \
	RORW	$(32-shift), a; \
	MOVW	c, R9; \
	ADDW	b, a