// CHECK: %[[VAL_0:.*]] = "tf._FusedConv2D"(%arg2, %arg1, %arg0) <{data_format = "NHWC", dilations = [1, 1, 1, 1], epsilon = 0.000000e+00 : f32, explicit_paddings = [], fused_ops = ["BiasAdd", "Elu"], num_args = 1 : i64, operandSegmentSizes = array<i32: 1, 1, 1, 0>, padding = "SAME", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true}> {TArgs = [f32]} : (tensor<8x32x32x3xf32>, tensor<1x1x3x128xf32>, tensor<128xf32>) -> tensor<*xf32>

  def testWithKnownKernel(self):

    @def_function.function
    def biasd_dense_elu(x, y, z):
      dot = gen_composite_ops.my_biased_dense(x, y, z)
      return nn_ops.elu(dot)  # with known kernel, should not expand.

    t1 = constant_op.constant([[1.0, 2.0], [3.0, 4.0]])
    t2 = constant_op.constant([[1.0, 2.0], [3.0, 4.0]])
    t3 = constant_op.constant([[-10.0, -10.0], [-10.0, -10.0]])

    self.assertAllEqual(sq.numpy().reshape(-1), [0, 0, 5, 12])

  def testWithKnownKernel(self):

    def biasd_dense_elu(x, y, z):
      dot = gen_composite_ops.my_biased_dense(x, y, z)
      return nn_ops.elu(dot)  # with known kernel, should not expand.

    t1 = constant_op.constant([[1.0, 2.0], [3.0, 4.0]])
    t2 = constant_op.constant([[1.0, 2.0], [3.0, 4.0]])
    t3 = constant_op.constant([[-10.0, -10.0], [-10.0, -10.0]])

}

tfr.func @tf__my_biased_dense_(!tfr.tensor<T>, !tfr.tensor<T>, !tfr.tensor<T>,
    !tfr.attr{tfr.name="act", tfr.default=""}) -> !tfr.tensor attributes {T}

// This is a wong decomposition and used to verify that tf.Elu isn't decomposed
// since its kernel has been registered.
tfr.func @tf__elu_(%input: !tfr.tensor) -> !tfr.tensor {
  tfr.return %input : !tfr.tensor
}

// Translated from:
//
// REGISTER_OP("Add")

namespace tensorflow {
namespace {

using ops::AvgPool;
using ops::AvgPool3D;
using ops::BiasAdd;
using ops::Conv2D;
using ops::Conv2DBackpropInput;
using ops::DepthwiseConv2dNative;
using ops::Elu;
using ops::FractionalAvgPool;
using ops::FractionalMaxPool;
using ops::FusedBatchNormV3;
using ops::L2Loss;
using ops::LogSoftmax;
using ops::LRN;
using ops::MaxPool;
using ops::MaxPool3D;
using ops::MaxPoolV2;

  %scale = arith.constant 0.1 : f32
  %zp = arith.constant 42 : i64
  %elu_attr = tfr.constant "ELU" -> !tfr.attr
  %min, %max = "tfr.quant_act_range"(%elu_attr, %scale, %zp) : (!tfr.attr, f32, i64) -> (!tfr.tensor, !tfr.tensor)
  func.return %min, %max : !tfr.tensor, !tfr.tensor
// CHECK: %[[elu_attr:.*]] = tfr.constant "ELU" -> !tfr.attr
// CHECK: %[[min:.*]], %[[max:.*]] = tfr.quant_act_range(%[[elu_attr]]

def : Pat<(TF_OnesLikeOp AnyTensor:$arg),
          (MHLO_ConstantLike<"1"> $arg)>;

//===----------------------------------------------------------------------===//
// Elu op patterns.
//===----------------------------------------------------------------------===//

def : Pat<(TF_EluOp AnyTensor:$features),
          (MHLO_SelectOp
           (MHLO_CompareOp
              $features,

                     std::vector<Output>* grad_outputs) {
  auto dx = internal::EluGrad(scope, grad_inputs[0], op.output(0));
  grad_outputs->push_back(dx);
  return scope.status();
}
REGISTER_GRADIENT_OP("Elu", EluGradHelper);

Status SeluGradHelper(const Scope& scope, const Operation& op,
                      const std::vector<Output>& grad_inputs,
                      std::vector<Output>* grad_outputs) {

  MIRROR_PAD = 100,
  ABS = 101,
  SPLIT_V = 102,
  UNIQUE = 103,
  CEIL = 104,
  REVERSE_V2 = 105,
  ADD_N = 106,
  GATHER_ND = 107,
  COS = 108,
  WHERE = 109,
  RANK = 110,
  ELU = 111,
  REVERSE_SEQUENCE = 112,
  MATRIX_DIAG = 113,
  QUANTIZE = 114,
  MATRIX_SET_DIAG = 115,
  ROUND = 116,
  HARD_SWISH = 117,
  IF = 118,
  WHILE = 119,
  NON_MAX_SUPPRESSION_V4 = 120,

  MIRROR_PAD = 100,
  ABS = 101,
  SPLIT_V = 102,
  UNIQUE = 103,
  CEIL = 104,
  REVERSE_V2 = 105,
  ADD_N = 106,
  GATHER_ND = 107,
  COS = 108,
  WHERE = 109,
  RANK = 110,
  ELU = 111,
  REVERSE_SEQUENCE = 112,
  MATRIX_DIAG = 113,
  QUANTIZE = 114,
  MATRIX_SET_DIAG = 115,
  ROUND = 116,
  HARD_SWISH = 117,
  IF = 118,
  WHILE = 119,
  NON_MAX_SUPPRESSION_V4 = 120,