// come in on the m->g0 stack already. Or we might already
	// be on the m->gsignal stack.
	get_tls(CX)
	MOVQ	g(CX), DI
	CMPQ	DI, $0
	JEQ	nosave
	MOVQ	g_m(DI), R8
	MOVQ	m_gsignal(R8), SI
	CMPQ	DI, SI
	JEQ	nosave
	MOVQ	m_g0(R8), SI
	CMPQ	DI, SI
	JEQ	nosave

	// Switch to system stack.
	// The original frame pointer is stored in BP,
	// which is useful for stack unwinding.

// so that regmask stays within int64
// Be careful when hand coding regmasks.
var regNamesARM64 = []string{
	"R0",
	"R1",
	"R2",
	"R3",
	"R4",
	"R5",
	"R6",
	"R7",
	"R8",
	"R9",
	"R10",
	"R11",
	"R12",
	"R13",
	"R14",
	"R15",
	"R16",
	"R17",
	"R18", // platform register, not used
	"R19",
	"R20",
	"R21",
	"R22",
	"R23",
	"R24",
	"R25",
	"R26",

	// Index registers needed for BR movs
	MOVD	$8, R9
	MOVD	$16, R10
	MOVD	$24, R14

	MOVDBR	(R0)(R4), R5
	MOVDBR	(R9)(R4), R6
	MOVDBR	(R10)(R4), R7
	MOVDBR	(R14)(R4), R8

	MOVD	R8, 0(R3)
	MOVD	R7, 8(R3)
	MOVD	R6, 16(R3)
	MOVD	R5, 24(R3)

	RET

#define P3ptr   R3
#define P1ptr   R4
#define COUNT   R5

#define X1L    V0
#define X1H    V1