#define VMULT(x1, x2, out_low, out_hi) \
	VMULEUW x1, x2, TMP1; \
	VMULOUW x1, x2, TMP2; \
	VMRGEW TMP1, TMP2, out_hi; \
	VMRGOW TMP1, TMP2, out_low

//
// Vector multiply add word
//
//	VMALF  x0, x1, y, out_low
//	VMALHF x0, x1, y, out_hi
#define VMULT_ADD(x1, x2, y, one, out_low, out_hi) \
	VMULEUW  y, one, TMP2; \
	VMULOUW  y, one, TMP1; \
	VMULEUW  x1, x2, out_low; \
	VMULOUW  x1, x2, out_hi; \

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
	VSLDOI	$12, ZERO, TMP, TMP // vsldoi 6,0,6,12
	VXOR	IN0, TMP, IN0       // vxor 1,1,6
	VSLDOI	$12, ZERO, TMP, TMP // vsldoi 6,0,6,12
	VXOR	IN0, TMP, IN0       // vxor 1,1,6

	VSLDOI	$8, ZERO, IN1, STAGE  // vsldoi 7,0,2,8

	modes := []string{"copy", "reverse", "reverse1", "reverse2", "sort", "dither"}
	var tmp1, tmp2 [1025]int
	for _, n := range sizes {
		for m := 1; m < 2*n; m *= 2 {
			for dist := 0; dist < _NDist; dist++ {
				j := 0
				k := 1
				data := tmp1[0:n]
				for i := 0; i < n; i++ {
					switch dist {
					case _Sawtooth:
						data[i] = i % m
					case _Rand:

#  else
#define NEEDS_ESPERM
#define P8_LXVB16X(RA,RB,VT) \
	LXVD2X  (RA+RB), VT \
	VPERM	VT, VT, ESPERM, VT

#define P8_STXVB16X(VS,RA,RB) \
	VPERM	VS, VS, ESPERM, TMP2; \
	STXVD2X TMP2, (RA+RB)

#  endif
#else
#define P8_LXVB16X(RA,RB,VT) \
	LXVD2X  (RA+RB), VT

#define P8_STXVB16X(VS,RA,RB) \
	STXVD2X VS, (RA+RB)

#endif

#define MASK_PTR   R8

    @GwtIncompatible // used only from other GwtIncompatible code
    @Override
    void setBits(BitSet table) {
      BitSet tmp1 = new BitSet();
      first.setBits(tmp1);
      BitSet tmp2 = new BitSet();
      second.setBits(tmp2);
      tmp1.and(tmp2);
      table.or(tmp1);
    }

    @Override
    public String toString() {
      return "CharMatcher.and(" + first + ", " + second + ")";
    }

    @GwtIncompatible // used only from other GwtIncompatible code
    @Override
    void setBits(BitSet table) {
      BitSet tmp1 = new BitSet();
      first.setBits(tmp1);
      BitSet tmp2 = new BitSet();
      second.setBits(tmp2);
      tmp1.and(tmp2);
      table.or(tmp1);
    }

    @Override
    public String toString() {
      return "CharMatcher.and(" + first + ", " + second + ")";
    }

// Restrict this optimization to the stack, to avoid duplicating loads from the heap;
// see CL 145208 for discussion.
(Move {t1} [s] dst tmp1 midmem:(Move {t2} [s] tmp2 src _))
	&& t1.Compare(t2) == types.CMPeq
	&& isSamePtr(tmp1, tmp2)
	&& isStackPtr(src) && !isVolatile(src)
	&& disjoint(src, s, tmp2, s)
	&& (disjoint(src, s, dst, s) || isInlinableMemmove(dst, src, s, config))
	=> (Move {t1} [s] dst src midmem)

// The caller should order the right-hand side of the assignment before calling order.as2func.
// It rewrites,
//
//	a, b, a = ...
//
// as
//
//	tmp1, tmp2, tmp3 = ...
//	a, b, a = tmp1, tmp2, tmp3
//
// This is necessary to ensure left to right assignment order.
func (o *orderState) as2func(n *ir.AssignListStmt) {
	results := n.Rhs[0].Type()

		v3.AddArg3(v4, d3, v5)
		v1.AddArg3(v2, d2, v3)
		v.AddArg3(v0, d1, v1)
		return true
	}
	// match: (Move {t1} [s] dst tmp1 midmem:(Move {t2} [s] tmp2 src _))
	// cond: t1.Compare(t2) == types.CMPeq && isSamePtr(tmp1, tmp2) && isStackPtr(src) && !isVolatile(src) && disjoint(src, s, tmp2, s) && (disjoint(src, s, dst, s) || isInlinableMemmove(dst, src, s, config))
	// result: (Move {t1} [s] dst src midmem)
	for {

// RUN: odml_to_stablehlo %s --allow-tf=false -o /tmp/temp.mlir; [ -f /tmp/temp.mlir ]; [ -f /tmp/debug_stablehlo.mlir ]
// RUN: odml_to_stablehlo %s --allow-tf=true -o /tmp/temp2.mlir; [ -f /tmp/temp2.mlir ]

module attributes {tf.versions = {bad_consumers = [], min_consumer = 12 : i32, producer = 975 : i32}, tf_saved_model.semantics}  {