case IsInf(x, 0):
		if IsInf(x, -1) {
			return Pow(1/x, -y) // Pow(-0, -y)
		}
		switch {
		case y < 0:
			return 0
		case y > 0:
			return Inf(1)
		}
	case y == 0.5:
		return Sqrt(x)
	case y == -0.5:
		return 1 / Sqrt(x)
	}

	yi, yf := Modf(Abs(y))
	if yf != 0 && x < 0 {
		return NaN()
	}
	if yi >= 1<<63 {
		// yi is a large even int that will lead to overflow (or underflow to 0)

	const width = 130
	const height = 50

	im := image.NewGray(image.Rectangle{Max: image.Point{X: width, Y: height}})
	for x := 0; x < width; x++ {
		for y := 0; y < height; y++ {
			dist := math.Sqrt(math.Pow(float64(x-width/2), 2)/3+math.Pow(float64(y-height/2), 2)) / (height / 1.5) * 255
			var gray uint8
			if dist > 255 {
				gray = 255
			} else {
				gray = uint8(dist)
			}

@app.route('/hello')
def hello():
    version = os.environ.get('SERVICE_VERSION')

    # do some cpu intensive computation
    x = 0.0001
    for i in range(0, 1000000):
        x = x + math.sqrt(x)

    return 'Hello version: %s, instance: %s\n' % (version, os.environ.get('HOSTNAME'))


@app.route('/health')
def health():
    return 'Helloworld is healthy', 200


if __name__ == "__main__":

//                      | R - (erf(x)-x)/x | <= 2
//
//
//         Remark. The formula is derived by noting
//          erf(x) = (2/sqrt(pi))*(x - x**3/3 + x**5/10 - x**7/42 + ....)
//         and that
//          2/sqrt(pi) = 1.128379167095512573896158903121545171688
//         is close to one. The interval is chosen because the fix
//         point of erf(x) is near 0.6174 (i.e., erf(x)=x when x is

	MOVQ    $~(1<<63), AX
	ANDQ    AX, BX // p = |p|
	MOVQ    q+8(FP), CX
	ANDQ    AX, CX // q = |q|
	MOVQ    $PosInf, AX
	CMPQ    AX, BX
	JLE     isInfOrNaN
	CMPQ    AX, CX
	JLE     isInfOrNaN
	// hypot = max * sqrt(1 + (min/max)**2)
	MOVQ    BX, X0
	MOVQ    CX, X1
	ORQ     CX, BX
	JEQ     isZero
	MOVAPD  X0, X2
	MAXSD   X1, X0
	MINSD   X2, X1
	DIVSD   X0, X1
	MULSD   X1, X1
	ADDSD   $1.0, X1
	SQRTSD  X1, X1

			continue
		}
		// Sqrt(Conj(z)) == Conj(Sqrt(z))
		if f := Sqrt(Conj(v.in)); !cAlike(Conj(v.want), f) && !cAlike(v.in, Conj(v.in)) {
			t.Errorf("Sqrt(%g) = %g, want %g", Conj(v.in), f, Conj(v.want))
		}
	}
	for _, pt := range branchPoints {
		if f0, f1 := Sqrt(pt[0]), Sqrt(pt[1]); !cVeryclose(f0, f1) {
			t.Errorf("Sqrt(%g) not continuous, got %g want %g", pt[0], f0, f1)
		}
	}

  ZerosLike \
  Shape \
  ExpandDims \
  OnesLike

${generate} \
  --category=math \
  Mul \
  Conj \
  AddV2 \
  MatMul \
  Neg \
  Sum \
  Sub \
  Div \
  DivNoNan \
  Exp \
  Sqrt \
  SqrtGrad \
  Log1p

${generate} \
  --category=nn \
  SparseSoftmaxCrossEntropyWithLogits \
  ReluGrad \
  Relu \
  BiasAdd \
  BiasAddGrad

${generate} \
  --category=resource_variable \

  // CHECK-DAG: %[[EPS_BCAST:.+]] = mhlo.constant dense<1.001000e-05> : tensor<256xf32>
  // CHECK-DAG: %[[VARIANCE_EPS:.+]] = mhlo.add %[[VARIANCE]], %[[EPS_BCAST]] : tensor<256xf32>
  // CHECK-DAG: %[[STDDEV:.+]] = mhlo.sqrt %[[VARIANCE_EPS]] : tensor<256xf32>
  // CHECK-DAG: %[[STDDEV_BCAST:.+]] = "mhlo.broadcast_in_dim"(%[[STDDEV]]) <{broadcast_dimensions = dense<1> : tensor<1xi64>}> : (tensor<256xf32>) -> tensor<4x256xf32>

	}

	// Expect a uniform distribution of byte values, which lie in [0, 255].
	var (
		mean       = 255.0 / 2
		stddev     = 256.0 / math.Sqrt(12.0)
		errorScale = stddev / math.Sqrt(float64(n))
	)

	expected := &statsResults{mean, stddev, 0.10 * errorScale, 0.08 * errorScale}

	// Cast bytes as floats to use the common distribution-validity checks.

 private:
  AbstractTensorHandlePtr exp_;
};

class SqrtGradientFunction : public GradientFunction {
 public:
  explicit SqrtGradientFunction(AbstractTensorHandle* sqrt) : sqrt_(sqrt) {
    sqrt->Ref();
  }
  Status Compute(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {