var R10 chan<- (<-chan (chan int)) = (chan <- <- chan chan int)(nil)
var R11 chan<- (chan<- (chan int)) = (chan <- chan <- chan int)(nil)
var R12 chan (chan<- (<-chan int)) = (chan chan <- <- chan int)(nil)
var R13 chan (chan<- (chan<- int)) = (chan chan <- chan <- int)(nil)

var F1 func() func() int
func F2() func() func() int

func (c *sigctxt) r10() uint64 { return c.regs().__gregs[_REG_R10] }
func (c *sigctxt) r11() uint64 { return c.regs().__gregs[_REG_R11] }
func (c *sigctxt) r12() uint64 { return c.regs().__gregs[_REG_R12] }
func (c *sigctxt) r13() uint64 { return c.regs().__gregs[_REG_R13] }
func (c *sigctxt) r14() uint64 { return c.regs().__gregs[_REG_R14] }
func (c *sigctxt) r15() uint64 { return c.regs().__gregs[_REG_R15] }

//go:nosplit
//go:nowritebarrierrec

// 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

	SYSCALL
	JCC	2(PC)
	NEGQ	AX
	MOVL	AX, ret+16(FP)
	RET

TEXT runtime·thr_start(SB),NOSPLIT,$0
	MOVQ	DI, R13 // m

	// set up FS to point at m->tls
	LEAQ	m_tls(R13), DI
	CALL	runtime·settls(SB)	// smashes DI

	// set up m, g
	get_tls(CX)
	MOVQ	m_g0(R13), DI
	MOVQ	R13, g_m(DI)
	MOVQ	DI, g(CX)

	CALL	runtime·stackcheck(SB)
	CALL	runtime·mstart(SB)

	MOVQ 0, AX			// crash (not reached)

func (c *sigctxt) r11() uint64 { return uint64(c.regs().gregs[_REG_R11]) }
func (c *sigctxt) r12() uint64 { return uint64(c.regs().gregs[_REG_R12]) }
func (c *sigctxt) r13() uint64 { return uint64(c.regs().gregs[_REG_R13]) }
func (c *sigctxt) r14() uint64 { return uint64(c.regs().gregs[_REG_R14]) }
func (c *sigctxt) r15() uint64 { return uint64(c.regs().gregs[_REG_R15]) }

//go:nosplit

	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)

// The zero value denotes R0, not the absence of a register.
type Reg uint8

const (
	R0 Reg = iota
	R1
	R2
	R3
	R4
	R5
	R6
	R7
	R8
	R9
	R10
	R11
	R12
	R13
	R14
	R15

	S0
	S1
	S2
	S3
	S4
	S5
	S6
	S7
	S8
	S9
	S10
	S11
	S12
	S13
	S14
	S15
	S16
	S17
	S18
	S19
	S20
	S21
	S22

	// GOT slot is filled by the dynamic linker with the offset of the thread-local
	// symbol from the thread pointer (R13)).
	R_POWER_TLS_IE

	// R_POWER_TLS marks an X-form instruction such as "ADD R3,R13,R4" as completing
	// a sequence of GOT-relative relocations to compute a TLS address. This can be
	// used by the system linker to to rewrite the GOT-relative TLS relocation into a

	MOVD	key+48(FP), R4
	MOVD	nonce+56(FP), R6
	MOVD	counter+64(FP), R7

	MOVD	$·constants(SB), R10
	MOVD	$·incRotMatrix(SB), R11

	MOVW	(R7), R20

	AND	$~255, R3, R13
	ADD	R2, R13, R12 // R12 for block end
	AND	$255, R3, R13
loop:
	MOVD	$NUM_ROUNDS, R21
	VLD1	(R11), [V30.S4, V31.S4]

	// load contants
	// VLD4R (R10), [V0.S4, V1.S4, V2.S4, V3.S4]
	WORD	$0x4D60E940

	// load keys

	MOVQ	(7*8)(DI), R10	// a7
	MOVQ	(8*8)(DI), R11	// a8
	MOVQ	(9*8)(DI), R12	// a9
	MOVQ	(10*8)(DI), R13	// a10
	MOVQ	R10, (0*8)(SP)	// a7
	MOVQ	R11, (1*8)(SP)	// a8
	MOVQ	R12, (2*8)(SP)	// a9
	MOVQ	R13, (3*8)(SP)	// a10
	MOVQ	(0*8)(DI), R11	// fn
	MOVQ	(2*8)(DI), SI	// a2
	MOVQ	(3*8)(DI), DX	// a3
	MOVQ	(4*8)(DI), CX	// a4
	MOVQ	(5*8)(DI), R8	// a5