// Compute the reflected value, used when x < 0.5:
    //
    //   lgamma(x) = log(pi) - lgamma(1-x) - log(abs(sin(pi * x))).
    //
    // (The abs is because lgamma is the log of the absolute value of the gamma
    // function.)
    //
    // We have to be careful when computing the final term above. gamma(x) goes
    // to +/-inf at every integer x < 0, and this is controlled by the

  %0 = "tf.Round"(%arg0) : (tensor<*xf32>) -> tensor<*xf32>
  func.return %0 : tensor<*xf32>
}

// CHECK-LABEL: func @lgamma
func.func @lgamma(%arg0: tensor<4xf32>) -> tensor<4xf32> {
  // The lowering for lgamma is complicated, which makes it awkward to write a
  // complete test for it here. Instead we test that Lgamma is at least being
  // lowered here and rely on UnaryOpsTest.testFloatOps and other TensorFlow

  Scope grad_scope = scope.WithControlDependencies(grad);
  auto x = ConjugateHelper(grad_scope, op.input(0));
  auto dx = Mul(grad_scope, grad, Digamma(grad_scope, x));
  grad_outputs->push_back(dx);
  return grad_scope.status();
}
REGISTER_GRADIENT_OP("Lgamma", LgammaGrad);

Status MinOrMaxGrad(const Scope& scope, const Operation& op,
                    const std::vector<Output>& grad_inputs,

            "Rsqrt", "Sigmoid", "Sign", "Sinh", "Softplus", "Softsign", "Sqrt",
            "Square", "Tan", "Tanh", "Real", "Imag", "Erf", "Erfc", "Erfinv",
            "Lgamma", "Digamma",
            // Binary
            "Add", "AddV2", "Sub", "Mul", "Div", "Atan2", "Complex", "DivNoNan",
            "MulNoNan", "FloorDiv", "Xlogy", "Xlog1py", "Xdivy", "FloorMod",

    %t:tensor<1x10x10x3xf32>, %m:tensor<3xf32>, %v:tensor<3xf32>, %beta:tensor<3xf32>, %gamma:tensor<3xf32>) -> (tensor<1x10x10x3xf32>) {
  %0 = "tf.BatchNormWithGlobalNormalization"(%t, %m, %v, %beta, %gamma) {T = "tfdtype$DT_FLOAT", variance_epsilon = 0.001 : f32, scale_after_normalization = false} : (tensor<1x10x10x3xf32>, tensor<3xf32>, tensor<3xf32>, tensor<3xf32>, tensor<3xf32>) -> (tensor<1x10x10x3xf32>)

If "scale_after_normalization" is true, this tensor will be multiplied
with the normalized tensor.}]>:$gamma,

    F32Attr:$variance_epsilon,
    BoolAttr:$scale_after_normalization
  );

  let results = (outs

?u&kir?ppihc?st???onihsnihs?ufuras??uaru??yru!koh??zimihs!ok?????nu,?g!iti,oyh!.&a&bmat?dnas?gusak?k&at?o&oyot?y??uzarakat??m&ayasas?irah??wa&g&ani?okak??k&i&hci?mak??oy???yi&hsa?monihsin???i&asak?hs&aka?i&at?nawak???j&awa!imanim??emih??k&a&goa?s&agama?ukuf??wihsin??i&hsog?m???mati?oia?rogimak??n&annas?esnonihs??o&gasa!kat??ka?n&ikat?o?ustat??rihsay?sihs?tomus?yas??u&bay?gnihs?????hih,konip,l&bs,ik,?mol,nagan!.&a&bukah?d&a&w?yim??e&ki?u??ii??k&a&s&ay?uki??zus??ihsoo?ousay??m&ay&akat?ii??i&hsukuf...

?u&kir?ppihc?st???onihsnihs?ufuras??uaru??yru!koh??zimihs!ok?????nu,?g!iti,oyh!.&a&bmat?dnas?gusak?k&at?o&oyot?y??uzarakat??m&ayasas?irah??wa&g&ani?okak??k&i&hci?mak??oy???yi&hsa?monihsin???i&asak?hs&aka?i&at?nawak???j&awa!imanim??emih??k&a&goa?s&agama?ukuf??wihsin??i&hsog?m???mati?oia?rogimak??n&annas?esnonihs??o&gasa!kat??ka?n&ikat?o?ustat??rihsay?sihs?tomus?yas??u&bay?gnihs?????hih,konip,l&bs,ik,?mol,nagan!.&a&bukah?d&a&w?yim??e&ki?u??ii??k&a&s&ay?uki??zus??ihsoo?ousay??m&ay&akat?ii??i&hsukuf...