i := 0
	for n != 0 {
		// amd64/v1,amd64/v2:"BSFL",-"BTSQ"
		// amd64/v3:"TZCNTL"
		i += bits.TrailingZeros32(n)
		n &= n - 1
	}
	return i
}

func IterateBits16(n uint16) int {
	i := 0
	for n != 0 {
		// amd64/v1,amd64/v2:"BSFL",-"BTSL"
		// amd64/v3:"TZCNTL"
		// arm64:"RBITW","CLZW",-"ORR"
		i += bits.TrailingZeros16(n)
		n &= n - 1
	}
	return i
}

	"SUBSS",
	"SUBW",
	"SWAPGS",
	"SYSCALL",
	"SYSENTER",
	"SYSENTER64",
	"SYSEXIT",
	"SYSEXIT64",
	"SYSRET",
	"TESTB",
	"TESTL",
	"TESTQ",
	"TESTW",
	"TPAUSE",
	"TZCNTL",
	"TZCNTQ",
	"TZCNTW",
	"UCOMISD",
	"UCOMISS",
	"UD1",
	"UD2",
	"UMWAIT",
	"UNPCKHPD",
	"UNPCKHPS",
	"UNPCKLPD",
	"UNPCKLPS",
	"UMONITOR",
	"V4FMADDPS",
	"V4FMADDSS",
	"V4FNMADDPS",

		"pmovsxbw", "pmovsxbd", "pmovsxbq", "pmovsxwd", "pmovsxwq", "pmovsxdq",
		"pblendvb",
	},
	"fma":   {"vfmadd231sd"},
	"movbe": {"movbeqq", "movbeq", "movbell", "movbel", "movbe"},
	"lzcnt": {"lzcntq", "lzcntl", "lzcnt"},
}

// Test to use POPCNT instruction, if available
func TestPopCnt(t *testing.T) {
	for _, tt := range []struct {
		x    uint64
		want int
	}{
		{0b00001111, 4},

(Ctz8  x) => (BSFL (ORLconst <typ.UInt32> [1<<8 ] x))

(Ctz64NonZero x) && buildcfg.GOAMD64 >= 3 => (TZCNTQ x)
(Ctz32NonZero x) && buildcfg.GOAMD64 >= 3 => (TZCNTL x)
(Ctz16NonZero x) && buildcfg.GOAMD64 >= 3 => (TZCNTL x)
(Ctz8NonZero  x) && buildcfg.GOAMD64 >= 3 => (TZCNTL x)
(Ctz64NonZero x) && buildcfg.GOAMD64 <  3 => (Select0 (BSFQ x))
(Ctz32NonZero x) && buildcfg.GOAMD64 <  3 => (BSFL x)

		// count the number of trailing zero bits, prefer TZCNTQ over BSFQ, as TZCNTQ(0)==64
		// and BSFQ(0) is undefined. Same for TZCNTL(0)==32
		{name: "TZCNTQ", argLength: 1, reg: gp11, asm: "TZCNTQ", clobberFlags: true},
		{name: "TZCNTL", argLength: 1, reg: gp11, asm: "TZCNTL", clobberFlags: true},

		// CPUID feature: LZCNT.
		// count the number of leading zero bits.

	TZCNTL (BX), DX                         // f30fbc13
	TZCNTL (R11), DX                        // f3410fbc13
	TZCNTL DX, DX                           // f30fbcd2
	TZCNTL R11, DX                          // f3410fbcd3
	TZCNTL (BX), R11                        // f3440fbc1b
	TZCNTL (R11), R11                       // f3450fbc1b
	TZCNTL DX, R11                          // f3440fbcda

// systems by fcntl_linux_32bit.go to be SYS_FCNTL64.
var fcntl64Syscall uintptr = SYS_FCNTL

func fcntl(fd int, cmd, arg int) (int, error) {
	valptr, _, errno := Syscall(fcntl64Syscall, uintptr(fd), uintptr(cmd), uintptr(arg))
	var err error
	if errno != 0 {
		err = errno
	}
	return int(valptr), err
}

// FcntlInt performs a fcntl syscall on fd with the provided command and argument.
func FcntlInt(fd uintptr, cmd, arg int) (int, error) {

	ASUBQ
	ASUBSD
	ASUBSS
	ASUBW
	ASWAPGS
	ASYSCALL
	ASYSENTER
	ASYSENTER64
	ASYSEXIT
	ASYSEXIT64
	ASYSRET
	ATESTB
	ATESTL
	ATESTQ
	ATESTW
	ATPAUSE
	ATZCNTL
	ATZCNTQ
	ATZCNTW
	AUCOMISD
	AUCOMISS
	AUD1
	AUD2
	AUMWAIT
	AUNPCKHPD
	AUNPCKHPS
	AUNPCKLPD
	AUNPCKLPS
	AUMONITOR
	AV4FMADDPS
	AV4FMADDSS
	AV4FNMADDPS
	AV4FNMADDSS
	AVADDPD

package unix

import (
	"syscall"
	_ "unsafe" // for go:linkname
)

// Implemented in the runtime package.
//
//go:linkname fcntl runtime.fcntl
func fcntl(fd int32, cmd int32, arg int32) (int32, int32)

func Fcntl(fd int, cmd int, arg int) (int, error) {
	val, errno := fcntl(int32(fd), int32(cmd), int32(arg))
	if val == -1 {
		return int(val), syscall.Errno(errno)
	}
	return int(val), nil

package unix

import "unsafe"

// FcntlInt performs a fcntl syscall on fd with the provided command and argument.
func FcntlInt(fd uintptr, cmd, arg int) (int, error) {
	return fcntl(int(fd), cmd, arg)
}

// FcntlFlock performs a fcntl syscall for the F_GETLK, F_SETLK or F_SETLKW command.
func FcntlFlock(fd uintptr, cmd int, lk *Flock_t) error {
	_, err := fcntl(int(fd), cmd, int(uintptr(unsafe.Pointer(lk))))
	return err
}