attr {
    key: "dtype"
    value {
      type: DT_FLOAT
    }
  }
}
node {
  name: "training/SGD/gradients/bn5c_branch2c_1/FusedBatchNorm_grad/FusedBatchNormGrad"
  op: "FusedBatchNormGrad"
  input: "training/SGD/gradients/activation_48_1/Relu_grad/ReluGrad"
  input: "res5c_branch2c_1/Conv2D"
  input: "bn5c_branch2c_1/ReadVariableOp"
  input: "bn5c_branch2c_1/FusedBatchNorm:3"

 AssignSubVariableOp 106
 BiasAdd 1
 BiasAddGrad 1
 Cast 3
 Const 357
 Conv2D 53
 Conv2DBackpropFilter 53
 Conv2DBackpropInput 52
 DivNoNan 1
 Equal 1
 FusedBatchNorm 53
 FusedBatchNormGrad 53
 Identity 2
 MatMul 3
 MaxPool 1
 MaxPoolGrad 1
 Mean 1
 Mul 164
 Pad 1
 ReadVariableOp 646
 Relu 49
 ReluGrad 49
 Reshape 2
 ResourceApplyKerasMomentum 161
 ShapeN 50
 Softmax 1

  auto y = FusedBatchNormV3(scope_, x, scale, offset, mean, var, attrs);
  RunTest({x, scale, offset}, {shape, scale_shape, scale_shape}, {y.y},
          {shape});
}

INSTANTIATE_TEST_SUITE_P(
    FusedBatchNormGrad, FusedBatchNormGradTest,
    ::testing::Combine(::testing::Bool(), ::testing::Bool(),
                       ::testing::Values(TensorShape({2, 3, 4, 5}),

  grad_outputs->push_back(dx);
  return scope.status();
}
REGISTER_GRADIENT_OP("FractionalMaxPool", FractionalMaxPoolGradHelper);

// Templated constructor for FusedBatchNormGrad[..]::Attrs.
template <typename T>
T FusedBatchNormGradAttrs(float epsilon, StringPiece data_format,
                          bool is_training) {
  T result;
  result.epsilon_ = epsilon;

          {"REDUCEWINDOWPW", {"BiasAddGrad", "LRN", "LRNGrad"}},
          {"BN",
           {"FusedBatchNorm", "FusedBatchNormV2", "FusedBatchNormV3",
            "_FusedBatchNormEx", "FusedBatchNormGrad", "FusedBatchNormGradV2",
            "FusedBatchNormGradV3"}},
          {"Conv", {"_FusedConv2D"}},
          {"SORT", {"TopKV2"}},  // XLA version much faster then TF version.
          {"MISC",

  // CHECK-NEXT: %[[x_backprop:.*]] = mhlo.convert %[[mul3]] : tensor<8x8x8x8xf32>
  // CHECK-NEXT: return %[[x_backprop]] : tensor<8x8x8x8xf32>

    return success();
  }
};

using ConvertRFFTOp = ConvertFFTOp<TF::RFFTOp>;
using ConvertIRFFTOp = ConvertFFTOp<TF::IRFFTOp>;

// The base class to convert TensorFlow FusedBatchNormGrad*Op to HLO
// BatchNormGradOp for training and a sequence of binary ops for inference.
// TODO(b/145536565): move to legalize_tf_patterns.td if it applies.
template <typename FusedBatchNormGradOpT>

  );

  TF_DerivedOperandTypeAttr T = TF_DerivedOperandTypeAttr<0>;

  let hasCanonicalizer = 1;

  let hasVerifier = 1;
}

def TF_FusedBatchNormGradOp : TF_Op<"FusedBatchNormGrad", [Pure]> {
  let summary = "Gradient for batch normalization.";

  let description = [{
Note that the size of 4D Tensors are defined by either "NHWC" or "NCHW".