SHLQ   $0x0d, DI, SI
	SHLQ   $0x0d, R9, R8
	SHLQ   $0x0d, R11, R10
	SHLQ   $0x0d, R13, R12
	SHLQ   $0x0d, R15, R14
	ANDQ   AX, DI
	IMUL3Q $0x13, R14, R14
	ADDQ   R14, DI
	ANDQ   AX, R9
	ADDQ   SI, R9
	ANDQ   AX, R11
	ADDQ   R8, R11
	ANDQ   AX, R13
	ADDQ   R10, R13
	ANDQ   AX, R15
	ADDQ   R12, R15

	// Second reduction chain (carryPropagate)
	MOVQ   DI, SI
	SHRQ   $0x33, SI
	MOVQ   R9, R8

	var a uint16

	z &= 15

	// amd64:"ROLW",-"ANDQ"
	// riscv64: "OR","SLL","SRL",-"AND\t"
	a += x<<z | x>>(16-z)

	// amd64:"RORW",-"ANDQ"
	// riscv64: "OR","SLL","SRL",-"AND\t"
	a += x>>z | x<<(16-z)

	return a
}

func rot8nc(x uint8, z uint) uint8 {
	var a uint8

	z &= 7

	// amd64:"ROLB",-"ANDQ"
	// riscv64: "OR","SLL","SRL",-"AND\t"
	a += x<<z | x>>(8-z)

	// 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"

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

	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

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)