// the compiler word, the link editor word, and the TOC save word.
const MinFrameSize = _MinFrameSize

// StackAlign is the required alignment of the SP register.
// The stack must be at least word aligned, but some architectures require more.

	VMOVB	W0, -48(R4)
	VMOVB	W0, -32(R4)
	VMOVB	W0, -16(R4)
	JMP	done

no_msa:
	// if less than 8 bytes, do one byte at a time
	SGTU	$8, R2, R3
	BNE	R3, out

	// do one byte at a time until 8-aligned
	AND	$7, R1, R3
	BEQ	R3, words
	MOVB	R0, (R1)
	ADDV	$1, R1
	JMP	-4(PC)

words:
	// do 8 bytes at a time if there is room
	ADDV	$-7, R4, R2

	SGTU	R2, R1, R3
	BEQ	R3, out
	MOVV	R0, (R1)

	ifmam2 := &wireFormat{extOff: sizeofIfmaMsghdr2Darwin15, bodyOff: sizeofIfmaMsghdr2Darwin15}
	ifmam2.parse = ifmam2.parseInterfaceMulticastAddrMessage
	// Darwin kernels require 32-bit aligned access to routing facilities.
	return 4, map[int]*wireFormat{
		syscall.RTM_ADD:       rtm,
		syscall.RTM_DELETE:    rtm,
		syscall.RTM_CHANGE:    rtm,
		syscall.RTM_GET:       rtm,
		syscall.RTM_LOSING:    rtm,

	DMB	MB_ISHST

	STREXD	R2, (R1), R0	// stores R2 and R3
	CMP	$0, R0
	BNE	store64loop

	DMB	MB_ISH
	RET

// The following functions all panic if their address argument isn't
// 8-byte aligned. Since we're calling back into Go code to do this,
// we have to cooperate with stack unwinding. In the normal case, the
// functions tail-call into the appropriate implementation, which

// two's-complement way.
//
//go:nosplit
func (i *Int32) Add(delta int32) int32 {
	return Xaddint32(&i.value, delta)
}

// Int64 is an atomically accessed int64 value.
//
// 8-byte aligned on all platforms, unlike a regular int64.
//
// An Int64 must not be copied.
type Int64 struct {
	noCopy noCopy
	_      align64
	value  int64
}

// Load accesses and returns the value atomically.
//

	MOVV	$1, R1
	MOVB	R1, ret+24(FP)
	RET
byte_test:
	MOVBU	(R1), R6
	ADDV	$1, R1
	MOVBU	(R2), R7
	ADDV	$1, R2
	BEQ	R6, R7, byte_loop
	JMP	not_eq

chunk_entry:
	// make sure both a and b are aligned
	OR	R1, R2, R9
	AND	$0x7, R9
	BNE	R0, R9, byte_loop
	JMP	chunk_loop_1

chunk_loop:
	// chunk size is 16
	SGTU	$16, R3, R8
	BNE	R0, R8, chunk_tail_8
chunk_loop_1:
	MOVV	(R1), R6
	MOVV	(R2), R7

    // The cache is effectively disabled, so we pass the read through to the
    // fetcher without breaking it up into blocks.
    return block_fetcher_(filename, offset, n, buffer, status);
  }
  // Calculate the block-aligned start and end of the read.
  size_t start = block_size_ * (offset / block_size_);
  size_t finish = block_size_ * ((offset + n) / block_size_);
  if (finish < offset + n) {
    finish += block_size_;
  }

_b4align:				/* align destination on 4 */
	AND.S	$3, TE, TMP
	BEQ	_b4aligned

	MOVBU.W	-1(FROM), TMP	/* pre-indexed */
	MOVBU.W	TMP, -1(TE)	/* pre-indexed */
	B	_b4align

_b4aligned:				/* is source now aligned? */
	AND.S	$3, FROM, TMP
	BNE	_bunaligned

	ADD	$31, TS, TMP	/* do 32-byte chunks if possible */
	MOVW	TS, savedts-4(SP)
_b32loop:
	CMP	TMP, TE
	BLS	_b4tail

	MOVM.DB.W (FROM), [R0-R7]

	MOVV	ptr+0(FP), R4
	MOVBU	val+8(FP), R5
	// Align ptr down to 4 bytes so we can use 32-bit load/store.
	MOVV	$~3, R6
	AND	R4, R6
	// R7 = ((ptr & 3) * 8)
	AND	$3, R4, R7
	SLLV	$3, R7
	// Shift val for aligned ptr. R5 = val << R4
	SLLV	R7, R5

	DBAR
	LL	(R6), R7
	OR	R5, R7
	SC	R7, (R6)
	BEQ	R7, -4(PC)
	DBAR
	RET

// void	And8(byte volatile*, byte);
TEXT ·And8(SB), NOSPLIT, $0-9

// check is negligible since it is only required for large copies.
//
// Large copies use a software pipelined loop processing 64 bytes per iteration.
// The destination pointer is 16-byte aligned to minimize unaligned accesses.
// The loop tail is handled by always copying 64 bytes from the end.

// func memmove(to, from unsafe.Pointer, n uintptr)
TEXT runtime·memmove<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-24