module_str = self._extract_first_xla_call_module_op(
        self._output_saved_model_path
    )

    # Tests that the output graph contains multiply for symmetric
    # dequantization.
    self.assertTrue(re.search('stablehlo.multiply', module_str))
    # Tests that the output graph contains float dot_general.
    self.assertTrue(
        re.search('stablehlo.dot_general.*xf32>.*xf32>.*xf32>', module_str)

// %13 = stablehlo.subtract %7, %12  // q1 * q2 - q2 * z1
// %14 = stablehlo.constant  // Merged scale s1 * s2, precalculated.
// %15 = stablehlo.broadcast_in_dim %14
// %16 = stablehlo.multiply %13 %15  // r3 = s1 s2 (q1 q2 - q2 z1)
//
// The following quant -> dequant pattern is a no-op, but is required to
// retrieve the quantization parameters for the output tensor.
//

// equivalent function from the corresponding s390x
// instruction for vector multiply high, low, and add,
// since there aren't exact equivalent instructions.
// The corresponding s390x instructions appear in the
// comments.
// Implementation for big endian would have to be
// investigated, I think it would be different.
//
//
// Vector multiply word
//
//	VMLF  x0, x1, out_low
//	VMLHF x0, x1, out_hi

		{name: "LoweredNilCheck", argLength: 2, reg: regInfo{inputs: []regMask{ptrsp}}, clobberFlags: true, nilCheck: true, faultOnNilArg0: true},
		// Round ops to block fused-multiply-add extraction.
		{name: "LoweredRound32F", argLength: 1, reg: fp11, resultInArg0: true, zeroWidth: true},
		{name: "LoweredRound64F", argLength: 1, reg: fp11, resultInArg0: true, zeroWidth: true},

(Sub32F ...) => (FSUBS ...)
(Sub64F ...) => (FSUB ...)

(Min(32|64)F x y) && buildcfg.GOPPC64 >= 9 => (XSMINJDP x y)
(Max(32|64)F x y) && buildcfg.GOPPC64 >= 9 => (XSMAXJDP x y)

// Combine 64 bit integer multiply and adds
(ADD l:(MULLD x y) z) && buildcfg.GOPPC64 >= 9 && l.Uses == 1 && clobber(l) => (MADDLD x y z)

(Mod16 x y) => (Mod32 (SignExt16to32 x) (SignExt16to32 y))
(Mod16u x y) => (Mod32u (ZeroExt16to32 x) (ZeroExt16to32 y))

//go:noinline
func fnmadd(x, y, z float64) float64 {
	return FMA(-x, y, -z)
}

func TestFMANegativeArgs(t *testing.T) {
	// Some architectures have instructions for fused multiply-subtract and
	// also negated variants of fused multiply-add and subtract. This test
	// aims to check that the optimizations that generate those instructions
	// are applied correctly, if they exist.
	for _, c := range fmaC {

// and the result are assumed to be in big-endian representation (to
// match the semantics of Int.Bytes and Int.SetBytes).
func mulBytes(x, y []byte) []byte {
	z := make([]byte, len(x)+len(y))

	// multiply
	k0 := len(z) - 1
	for j := len(y) - 1; j >= 0; j-- {
		d := int(y[j])
		if d != 0 {
			k := k0
			carry := 0
			for i := len(x) - 1; i >= 0; i-- {
				t := int(z[k]) + int(x[i])*d + carry

		payloadBytes-- // encrypted ContentType
	}

	// Allow packet growth in arithmetic progression up to max.
	pkt := c.packetsSent
	c.packetsSent++
	if pkt > 1000 {
		return maxPlaintext // avoid overflow in multiply below
	}

	n := payloadBytes * int(pkt+1)
	if n > maxPlaintext {
		n = maxPlaintext
	}
	return n
}

func (c *Conn) write(data []byte) (int, error) {
	if c.buffering {

}{
	{
		Value: make(chan bool),
		Err:   "xml: unsupported type: chan bool",
		Kind:  reflect.Chan,
	},
	{
		Value: map[string]string{
			"question": "What do you get when you multiply six by nine?",
			"answer":   "42",
		},
		Err:  "xml: unsupported type: map[string]string",
		Kind: reflect.Map,
	},
	{
		Value: map[*Ship]bool{nil: false},

//
// To count the number of units in a [Duration], divide:
//
//	second := time.Second
//	fmt.Print(int64(second/time.Millisecond)) // prints 1000
//
// To convert an integer number of units to a Duration, multiply:
//
//	seconds := 10
//	fmt.Print(time.Duration(seconds)*time.Second) // prints 10s
const (
	Nanosecond  Duration = 1
	Microsecond          = 1000 * Nanosecond
	Millisecond          = 1000 * Microsecond