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

	print("r8   ", hex(c.r8()), "\t")
	print("r9   ", hex(c.r9()), "\n")
	print("r10  ", hex(c.r10()), "\t")
	print("r11  ", hex(c.r11()), "\n")
	print("r12  ", hex(c.r12()), "\t")
	print("r13  ", hex(c.r13()), "\n")
	print("r14  ", hex(c.r14()), "\t")
	print("r15  ", hex(c.r15()), "\n")
	print("r16  ", hex(c.r16()), "\t")
	print("r17  ", hex(c.r17()), "\n")
	print("r18  ", hex(c.r18()), "\t")

func (c *sigctxt) r10() uint64 { return c.regs().r10 }
func (c *sigctxt) r11() uint64 { return c.regs().r11 }
func (c *sigctxt) r12() uint64 { return c.regs().r12 }
func (c *sigctxt) r13() uint64 { return c.regs().r13 }
func (c *sigctxt) r14() uint64 { return c.regs().r14 }
func (c *sigctxt) r15() uint64 { return c.regs().r15 }

//go:nosplit
//go:nowritebarrierrec
func (c *sigctxt) rip() uint64 { return c.regs().rip }

	MOVV R1, -480(R3)
	SUBV $480, R3
	MOVV R4, 8(R3)
	MOVV R5, 16(R3)
	MOVV R6, 24(R3)
	MOVV R7, 32(R3)
	MOVV R8, 40(R3)
	MOVV R9, 48(R3)
	MOVV R10, 56(R3)
	MOVV R11, 64(R3)
	MOVV R12, 72(R3)
	MOVV R13, 80(R3)
	MOVV R14, 88(R3)
	MOVV R15, 96(R3)
	MOVV R16, 104(R3)
	MOVV R17, 112(R3)
	MOVV R18, 120(R3)
	MOVV R19, 128(R3)
	MOVV R20, 136(R3)
	MOVV R21, 144(R3)
	MOVV R23, 152(R3)
	MOVV R24, 160(R3)

	print("r8      ", hex(c.r8()), "\n")
	print("r9      ", hex(c.r9()), "\n")
	print("r10     ", hex(c.r10()), "\n")
	print("r11     ", hex(c.r11()), "\n")
	print("r12     ", hex(c.r12()), "\n")
	print("r13     ", hex(c.r13()), "\n")
	print("r14     ", hex(c.r14()), "\n")
	print("r15     ", hex(c.r15()), "\n")
	print("r16     ", hex(c.r16()), "\n")
	print("r17     ", hex(c.r17()), "\n")
	print("r18     ", hex(c.r18()), "\n")

	MOVL $495, AX		// lwp_create
	SYSCALL
	MOVL	AX, ret+8(FP)
	RET

TEXT runtime·lwp_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)

	return ptracePtr(PTRACE_SETREGS, pid, 0, unsafe.Pointer(regs))
}

// PtraceRegsAmd64 is the registers used by amd64 binaries.
type PtraceRegsAmd64 struct {
	R15      uint64
	R14      uint64
	R13      uint64
	R12      uint64
	Rbp      uint64
	Rbx      uint64
	R11      uint64
	R10      uint64
	R9       uint64
	R8       uint64
	Rax      uint64
	Rcx      uint64
	Rdx      uint64
	Rsi      uint64

	print("r8   ", hex(c.r8()), "\t")
	print("r9   ", hex(c.r9()), "\n")
	print("r10  ", hex(c.r10()), "\t")
	print("r11  ", hex(c.r11()), "\n")
	print("r12  ", hex(c.r12()), "\t")
	print("r13  ", hex(c.r13()), "\n")
	print("r14  ", hex(c.r14()), "\t")
	print("r15  ", hex(c.r15()), "\n")
	print("r16  ", hex(c.r16()), "\t")
	print("r17  ", hex(c.r17()), "\n")
	print("r18  ", hex(c.r18()), "\t")

	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; \

	MOVQ    $0, R8
	// Copy mp, gp, fn off parent stack for use by child.
	// Careful: Linux system call clobbers CX and R11.
	MOVQ	mp+16(FP), R13
	MOVQ	gp+24(FP), R9
	MOVQ	fn+32(FP), R12
	CMPQ	R13, $0    // m
	JEQ	nog1
	CMPQ	R9, $0    // g
	JEQ	nog1
	LEAQ	m_tls(R13), R8
#ifdef GOOS_android
	// Android stores the TLS offset in runtime·tls_g.
	SUBQ	runtime·tls_g(SB), R8
#else