JE	call	// no g; still on a system stack

	MOVQ	g_m(R14), R13
	// Switch to g0 stack.
	MOVQ	m_g0(R13), R10
	CMPQ	R10, R14
	JE	call	// already on g0

	MOVQ	(g_sched+gobuf_sp)(R10), SP
call:
	ANDQ	$~15, SP	// alignment for gcc ABI
	CALL	AX
	MOVQ	R12, SP

	// s390x:"LPDFR",-"LDEBR",-"LEDBR"     (no float64 conversion)
	return float32(math.Abs(float64(x)))
}

// Check that it's using integer registers
func copysign(a, b, c float64) {
	// amd64:"BTRQ\t[$]63","ANDQ","ORQ"
	// s390x:"CPSDR",-"MOVD"         (no integer load/store)
	// ppc64x:"FCPSGN"
	// riscv64:"FSGNJD"
	// wasm:"F64Copysign"
	sink64[0] = math.Copysign(a, b)

	// amd64:"BTSQ\t[$]63"

	// amd64:"SHLQ.*,.*,"
	shl := x<<s | y>>ŝ
	return shr, shl
}

func checkShiftToMask(u []uint64, s []int64) {
	// amd64:-"SHR",-"SHL","ANDQ"
	u[0] = u[0] >> 5 << 5
	// amd64:-"SAR",-"SHL","ANDQ"
	s[0] = s[0] >> 5 << 5
	// amd64:-"SHR",-"SHL","ANDQ"
	u[1] = u[1] << 5 >> 5

TEXT runtime·asmstdcall_trampoline<ABIInternal>(SB),NOSPLIT,$0
	MOVQ	AX, CX
	JMP	runtime·asmstdcall(SB)

// void runtime·asmstdcall(void *c);
TEXT runtime·asmstdcall(SB),NOSPLIT,$16
	MOVQ	SP, AX
	ANDQ	$~15, SP	// alignment as per Windows requirement
	MOVQ	AX, 8(SP)
	MOVQ	CX, 0(SP)	// asmcgocall will put first argument into CX.

	MOVQ	libcall_fn(CX), AX
	MOVQ	libcall_args(CX), SI
	MOVQ	libcall_n(CX), CX

		{name: "ANDQ", argLength: 2, reg: gp21, asm: "ANDQ", commutative: true, resultInArg0: true, clobberFlags: true},                                                 // arg0 & arg1

	CMPB ·support_adx(SB), $1
	JEQ adx
	MOVQ z+0(FP), R10
	MOVQ x+24(FP), R8
	MOVQ y+48(FP), R9
	MOVQ z_len+8(FP), R11
	MOVQ $0, BX		// i = 0
	MOVQ $0, CX		// c = 0
	MOVQ R11, R12
	ANDQ $-2, R12
	CMPQ R11, $2
	JAE A6
	JMP E6

A6:
	MOVQ (R8)(BX*8), AX
	MULQ R9
	ADDQ (R10)(BX*8), AX
	ADCQ $0, DX
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ DX, CX
	MOVQ AX, (R10)(BX*8)

	MOVQ 	AX, ret+16(FP)
	RET

// func And64(addr *uint64, v uint64) old uint64
TEXT ·And64(SB), NOSPLIT, $0-24
	MOVQ	ptr+0(FP), BX
	MOVQ	val+8(FP), CX
casloop:
	MOVQ 	CX, DX
	MOVQ	(BX), AX
	ANDQ	AX, DX
	LOCK
	CMPXCHGQ	DX, (BX)
	JNZ casloop
	MOVQ 	AX, ret+16(FP)
	RET

// func Anduintptr(addr *uintptr, v uintptr) old uintptr
TEXT ·Anduintptr(SB), NOSPLIT, $0-24
	JMP	·And64(SB)

	lo = r.lo
	SHLQ(Imm(64-51), r.lo, r.hi)
	r.lo, r.hi = nil, nil // make sure the uint128 is unusable
	return
}

// maskAndAdd sets r = r&mask + c*i.
func maskAndAdd(r, mask, c GPVirtual, i uint64) {
	ANDQ(mask, r)
	if i != 1 {
		IMUL3Q(Imm(i), c, c)
	}
	ADDQ(c, r)
}

func mustAddr(c Component) Op {
	b, err := c.Resolve()
	if err != nil {
		panic(err)
	}
	return b.Addr

	MOVQ	_cgo_sys_thread_create(SB), AX
	TESTQ	AX, AX
	JZ	nocgo

	// We're calling back to C.
	// Align stack per ELF ABI requirements.
	MOVQ	SP, BX  // Callee-save in C ABI
	ANDQ	$~15, SP
	MOVQ	$_rt0_amd64_lib_go(SB), DI
	MOVQ	$0, SI
	CALL	AX
	MOVQ	BX, SP
	JMP	restore

nocgo:
	ADJSP	$16
	MOVQ	$0x800000, 0(SP)		// stacksize
	MOVQ	$_rt0_amd64_lib_go(SB), AX
	MOVQ	AX, 8(SP)			// fn

func bitoff64(a, b uint64) (n uint64) {
	// amd64:"BTRQ"
	n += b &^ (1 << (a & 63))

	// amd64:"BTRQ\t[$]63"
	n += a &^ (1 << 63)

	// amd64:"BTRQ\t[$]60"
	n += a &^ (1 << 60)

	// amd64:"ANDQ\t[$]-2"
	n += a &^ (1 << 0)

	return n
}

func bitcompl64(a, b uint64) (n uint64) {
	// amd64:"BTCQ"
	n += b ^ (1 << (a & 63))

	// amd64:"BTCQ\t[$]63"
	n += a ^ (1 << 63)