ADD	$16, OUTENC, OUTENC
	ADD	$-16, OUTDEC, OUTDEC
	VSPLTISB	$8, KEY                  // vspltisb 3,8
	MOVD	CNT, CTR                         // mtctr 7
	VSUBUBM	MASK, KEY, MASK                  // vsububm 5,5,3

loop192:
	VPERM	IN1, IN1, MASK, KEY // vperm 3,2,2,5
	VSLDOI	$12, ZERO, IN0, TMP // vsldoi 6,0,1,12
	VCIPHERLAST	KEY, RCON, KEY      // vcipherlast 3,3,4

	VXOR	IN0, TMP, IN0       // vxor 1,1,6

    ReducedPrecisionSupport mask = ReducedPrecisionSupport::None;
    if (toco_flags.quantize_to_float16()) {
      mask |= ReducedPrecisionSupport::Float16Inference;
    }
    if (toco_flags.allow_bfloat16()) {
      mask |= ReducedPrecisionSupport::Bfloat16Inference;
    }
    if (toco_flags.accumulation_type() == toco::IODataType::FLOAT16) {
      mask |= ReducedPrecisionSupport::Float16Accumulation;

		// Element is small with pointer mask; array is still small.
		// Create direct pointer mask by turning each 1 bit in elem
		// into length 1 bits in larger mask.
		n := (array.PtrBytes/goarch.PtrSize + 7) / 8
		// Runtime needs pointer masks to be a multiple of uintptr in size.
		n = (n + goarch.PtrSize - 1) &^ (goarch.PtrSize - 1)
		mask := make([]byte, n)
		emitGCMask(mask, 0, typ, array.Len)
		array.GCData = &mask[0]

//
//go:nosplit
func setSignalstackSP(s *stackt, sp uintptr) {
	s.ss_sp = sp
}

//go:nosplit
//go:nowritebarrierrec
func sigaddset(mask *sigset, i int) {
	mask.__bits[(i-1)/32] |= 1 << ((uint32(i) - 1) & 31)
}

func sigdelset(mask *sigset, i int) {
	mask.__bits[(i-1)/32] &^= 1 << ((uint32(i) - 1) & 31)
}

//go:nosplit
func (c *sigctxt) fixsigcode(sig uint32) {
}

	// for VPERMXOR
	MOVD $consts<>+0xa0(SB), MASK
	MOVD $16, R20
	// V16
	LXVW4X (CONSTBASE)(R0), VS48
	ADD $80,CONSTBASE

	// Load key into V17,V18
	LXVW4X (KEY)(R0), VS49
	LXVW4X (KEY)(R8), VS50

	// Load CNT, NONCE into V19
	LXVW4X (CNT)(R0), VS51

	// Clear V27
	VXOR V27, V27, V27

	// V28
	LXVW4X (CONSTBASE)(R11), VS60

	// Load mask constants for VPERMXOR
	LXVW4X (MASK)(R0), V20

type WaitStatus uint32

const (
	mask  = 0x7F
	core  = 0x80
	shift = 8

	exited  = 0
	stopped = 0x7F
)

func (w WaitStatus) Exited() bool { return w&mask == exited }

func (w WaitStatus) ExitStatus() int {
	if w&mask != exited {
		return -1
	}
	return int(w >> shift)
}

func (w WaitStatus) Signaled() bool { return w&mask != stopped && w&mask != 0 }

}

func bitSetTest(x int) bool {
	// amd64:"ANDL\t[$]9, AX"
	// amd64:"CMPQ\tAX, [$]9"
	return x&9 == 9
}

// mask contiguous one bits
func cont1Mask64U(x uint64) uint64 {
	// s390x:"RISBGZ\t[$]16, [$]47, [$]0,"
	return x & 0x0000ffffffff0000
}

// mask contiguous zero bits
func cont0Mask64U(x uint64) uint64 {
	// s390x:"RISBGZ\t[$]48, [$]15, [$]0,"
	return x & 0xffff00000000ffff
}

            source[offset]);
      }

      @Override
      public void putLongLittleEndian(byte[] sink, int offset, long value) {
        long mask = 0xFFL;
        for (int i = 0; i < 8; mask <<= 8, i++) {
          sink[offset + i] = (byte) ((value & mask) >> (i * 8));
        }
      }
    }
  }

  static {
    LittleEndianBytes theGetter = JavaLittleEndianBytes.INSTANCE;
    try {
      /*

//	VCHAR          =  %x21-7E
func validHeaderValueByte(c byte) bool {
	// mask is a 128-bit bitmap with 1s for allowed bytes,
	// so that the byte c can be tested with a shift and an and.
	// If c >= 128, then 1<<c and 1<<(c-64) will both be zero.
	// Since this is the obs-text range, we invert the mask to
	// create a bitmap with 1s for disallowed bytes.
	const mask = 0 |
		(1<<(0x7f-0x21)-1)<<0x21 | // VCHAR: %x21-7E
		1<<0x20 | // SP: %x20

func (d *poolDequeue) unpack(ptrs uint64) (head, tail uint32) {
	const mask = 1<<dequeueBits - 1
	head = uint32((ptrs >> dequeueBits) & mask)
	tail = uint32(ptrs & mask)
	return
}

func (d *poolDequeue) pack(head, tail uint32) uint64 {
	const mask = 1<<dequeueBits - 1
	return (uint64(head) << dequeueBits) |
		uint64(tail&mask)
}

// pushHead adds val at the head of the queue. It returns false if the