MOVQ	g(R12), R14
	MOVQ	g_m(R14), R13
	MOVQ	m_g0(R13), R15
	CMPQ	R13, R15
	JEQ	noswitch	// branch if already on g0
	MOVQ	R15, g(R12)	// g = m->g0
	MOVQ	R15, R14	// set g register
	PUSHQ	RARG1	// func arg
	PUSHQ	RARG0	// func arg
	CALL	runtime·racecallback(SB)
	POPQ	R12
	POPQ	R12
	// All registers are smashed after Go code, reload.
	get_tls(R12)
	MOVQ	g(R12), R13
	MOVQ	g_m(R13), R13

	// load_g will clobber R0, Save R0
	MOVD	R0, R13
	load_g
	// restore R0
	MOVD	R13, R0
	MOVD	g_m(g), R13
	MOVD	m_g0(R13), R14
	CMP	R14, g
	BEQ	noswitch	// branch if already on g0
	MOVD	R14, g

	MOVD	R0, 8(RSP)	// func arg
	MOVD	R1, 16(RSP)	// func arg
	BL	runtime·racecallback(SB)

	// All registers are smashed after Go code, reload.
	MOVD	g_m(g), R13
	MOVD	m_curg(R13), g	// g = m->curg
ret:

	BEQ	aligned			// aligned detected - skip copy

	// Copy the unaligned source data into the aligned temporary buffer
	// memmove(to=4(R13), from=8(R13), n=12(R13)) - Corrupts all registers
	MOVW	$buf, Rtable	// to
	MOVW	$64, Rc0		// n
	MOVM.IB	[Rtable,Rdata,Rc0], (R13)
	BL	runtime·memmove(SB)

	// Point to the local aligned copy of the data
	MOVW	$buf, Rdata

aligned:
	// Point to the table of constants

	MOVD	dig+0(FP), R0
	MOVD	p+8(FP), R1
	MOVD	p_len+16(FP), R2
	AND	$~63, R2
	CBZ	R2, zero

	ADD	R1, R2, R21
	LDPW	(0*8)(R0), (R4, R5)
	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; \

	// R4 = a_base
	// R5 = a_len
	// R6 = b_base
	// R7 = b_len
	JMP	cmpbody<>(SB)

// On entry:
// R5 length of a
// R7 length of b
// R4 points to the start of a
// R6 points to the start of b
// R13 points to the return value (-1/0/1)
TEXT cmpbody<>(SB),NOSPLIT|NOFRAME,$0
	BEQ	R4, R6, samebytes // same start of a and b

	SGTU	R5, R7, R9
	BNE	R0, R9, r2_lt_r1
	MOVV	R5, R14
	JMP	entry
r2_lt_r1:

	LEAQ	(SI)(DX*1),	DI
	MOVL	(0*4)(BP),	AX
	MOVL	(1*4)(BP),	BX
	MOVL	(2*4)(BP),	CX
	MOVL	(3*4)(BP),	DX
	MOVL	$0xffffffff,	R11

	CMPQ	SI,		DI
	JEQ	end

loop:
	MOVL	AX,		R12
	MOVL	BX,		R13
	MOVL	CX,		R14
	MOVL	DX,		R15

	MOVL	(0*4)(SI),	R8
	MOVL	DX,		R9

#define ROUND1(a, b, c, d, index, const, shift) \
	XORL	c, R9; \
	ADDL	$const, a; \
	ADDL	R8, a; \
	ANDL	b, R9; \
	XORL	d, R9; \

	MOVQ	p_len+16(FP),	DX
	SHRQ	$6,		DX
	SHLQ	$6,		DX

	MOVQ	$K_XMM_AR<>(SB), R8

	MOVQ	DI, R9
	MOVQ	SI, R10
	LEAQ	64(SI), R13

	ADDQ	SI, DX
	ADDQ	$64, DX
	MOVQ	DX, R11

	CMPQ	R13, R11
	CMOVQCC	R8, R13

	VMOVDQU	BSWAP_SHUFB_CTL<>(SB), Y10

	CALC // RET is inside macros

DATA K_XMM_AR<>+0x00(SB)/4,$0x5a827999
DATA K_XMM_AR<>+0x04(SB)/4,$0x5a827999

	MOVW	Rt0, p_end	// pointer to end of data

	// Load up initial SHA-1 accumulator
	MOVW	dig+0(FP), Rt0
	MOVM.IA (Rt0), [Ra,Rb,Rc,Rd,Re]

loop:
	// Save registers at SP+4 onwards
	MOVM.IB [Ra,Rb,Rc,Rd,Re], (R13)

	MOVW	$w_buf, Rw
	MOVW	$0x5A827999, Rconst
	MOVW	$3, Rctr
loop1:	ROUND1(Ra, Rb, Rc, Rd, Re)
	ROUND1(Re, Ra, Rb, Rc, Rd)
	ROUND1(Rd, Re, Ra, Rb, Rc)
	ROUND1(Rc, Rd, Re, Ra, Rb)

// do need to set the SP delta back).

// Check if R1 is 8-byte aligned, panic if not.
// Clobbers R2.
#define CHECK_ALIGN \
	AND.S	$7, R1, R2 \
	BEQ 	4(PC) \
	MOVW.W	R14, -4(R13) /* prepare a real frame */ \
	BL	·panicUnaligned(SB) \
	ADD	$4, R13 /* compensate SP delta */

TEXT ·Cas64(SB),NOSPLIT,$-4-21
	NO_LOCAL_POINTERS
	MOVW	addr+0(FP), R1
	CHECK_ALIGN

#ifndef GOARM_7

	MOVD	R9, R0

	// Now on a scheduling stack (a pthread-created stack).
	// Save room for two of our pointers /*, plus 32 bytes of callee
	// save area that lives on the caller stack. */
	MOVD	RSP, R13
	SUB	$16, R13
	MOVD	R13, RSP
	MOVD	R4, 0(RSP)	// save old g on stack
	MOVD	(g_stack+stack_hi)(R4), R4
	SUB	R2, R4
	MOVD	R4, 8(RSP)	// save depth in old g stack (can't just save SP, as stack might be copied during a callback)