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)

	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)

	// 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

func (c *sigctxt) r10() uint64 { return c.regs().sc_regs[10] }
func (c *sigctxt) r11() uint64 { return c.regs().sc_regs[11] }
func (c *sigctxt) r12() uint64 { return c.regs().sc_regs[12] }
func (c *sigctxt) r13() uint64 { return c.regs().sc_regs[13] }
func (c *sigctxt) r14() uint64 { return c.regs().sc_regs[14] }
func (c *sigctxt) r15() uint64 { return c.regs().sc_regs[15] }
func (c *sigctxt) r16() uint64 { return c.regs().sc_regs[16] }

func (c *sigctxt) r10() uint64 { return c.regs().sc_regs[10] }
func (c *sigctxt) r11() uint64 { return c.regs().sc_regs[11] }
func (c *sigctxt) r12() uint64 { return c.regs().sc_regs[12] }
func (c *sigctxt) r13() uint64 { return c.regs().sc_regs[13] }
func (c *sigctxt) r14() uint64 { return c.regs().sc_regs[14] }
func (c *sigctxt) r15() uint64 { return c.regs().sc_regs[15] }
func (c *sigctxt) r16() uint64 { return c.regs().sc_regs[16] }

	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

	// faster than having the caller spill these.
	MOVQ	R12, 96(SP)
	MOVQ	R13, 104(SP)
retry:
	// TODO: Consider passing g.m.p in as an argument so they can be shared
	// across a sequence of write barriers.
	MOVQ	g_m(R14), R13
	MOVQ	m_p(R13), R13
	// Get current buffer write position.
	MOVQ	(p_wbBuf+wbBuf_next)(R13), R12	// original next position
	ADDQ	R11, R12			// new next position
	// Is the buffer full?

func (c *sigctxt) r10() uint64 { return c.regs().gpr[10] }
func (c *sigctxt) r11() uint64 { return c.regs().gpr[11] }
func (c *sigctxt) r12() uint64 { return c.regs().gpr[12] }
func (c *sigctxt) r13() uint64 { return c.regs().gpr[13] }
func (c *sigctxt) r14() uint64 { return c.regs().gpr[14] }
func (c *sigctxt) r15() uint64 { return c.regs().gpr[15] }
func (c *sigctxt) r16() uint64 { return c.regs().gpr[16] }

	print("r7   ", hex(r.x[7]), "\n")
	print("r8   ", hex(r.x[8]), "\n")
	print("r9   ", hex(r.x[9]), "\n")
	print("r10  ", hex(r.x[10]), "\n")
	print("r11  ", hex(r.x[11]), "\n")
	print("r12  ", hex(r.x[12]), "\n")
	print("r13  ", hex(r.x[13]), "\n")
	print("r14  ", hex(r.x[14]), "\n")
	print("r15  ", hex(r.x[15]), "\n")
	print("r16  ", hex(r.x[16]), "\n")
	print("r17  ", hex(r.x[17]), "\n")
	print("r18  ", hex(r.x[18]), "\n")