// CHECK-NEXT:         outputs: [ 5 ],
// CHECK-NEXT:         builtin_options_type: SVDFOptions,
// CHECK-NEXT:         builtin_options: {
// CHECK-NEXT:           rank: 2,
// CHECK-NEXT:           fused_activation_function: RELU
// CHECK-NEXT:         }
// CHECK-NEXT:       } ],
// CHECK-NEXT:       name: "main"
// CHECK-NEXT:     } ],
// CHECK-NEXT:     description: "MLIR Converted.",
// CHECK-NEXT:     buffers: [ {
// CHECK-EMPTY:

    # output shape: [-1, 28, 28, 32]
    conv1 = gen_mnist_ops.new_conv2d(x, self.weights['f1'], self.biases['b1'],
                                     1, 1, 1, 1, 'SAME', 'RELU')

    # Max pooling. The kernel size spec {ksize} also follows the layout of
    # the data. Here we have a pooling window of 2, and a stride of 2.
    # output shape: [-1, 14, 14, 32]

// CHECK-NEXT:         outputs: [ 5 ],
// CHECK-NEXT:         builtin_options_type: SVDFOptions,
// CHECK-NEXT:         builtin_options: {
// CHECK-NEXT:           rank: 2,
// CHECK-NEXT:           fused_activation_function: RELU
// CHECK-NEXT:         }
// CHECK-NEXT:       } ],
// CHECK-NEXT:       name: "main"
// CHECK-NEXT:     } ],
// CHECK-NEXT:     description: "MLIR Converted.",
// CHECK-NEXT:     buffers: [ {
// CHECK-EMPTY:

 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
 SparseSoftmaxCrossEntropyWithLogits 1
 Square 55
 Squeeze 1
 Sub 106
 Sum 57
 Tile 1

 Conv2D 53
 Conv2DBackpropFilter 53
 Conv2DBackpropInput 52
 Equal 1
 FusedBatchNormGradV2 53
 FusedBatchNormV2 53
 MatMul 3
 MaxPool 1
 MaxPoolGrad 1
 Mean 1
 Mul 218
 Pad 2
 ReadVariableOp 538
 Relu 49
 ReluGrad 49
 Reshape 2
 ResourceApplyKerasMomentum 161
 Slice 1
 Softmax 1
 SparseSoftmaxCrossEntropyWithLogits 1
 Squeeze 1
 Sum 1
 Tile 1
 Transpose 1
cluster 1 size 815
 AddN 1

  // CHECK: tfrt_fallback_async.executeop key(0) cost(100) device("cpu") "tf.Relu"(%{{.*}}) {T = f32} : 1
  %2 = tfrt_fallback_async.executeop key(0) cost(100) device("cpu") "tf.Relu"(%1) {T = f32} : 1
  // CHECK: tfrt_fallback_async.fallback_tensor_to_corert_tensorhandle

// CHECK-NEXT:      outputs: [ 5, 6, 7, 8 ],
// CHECK-NEXT:      builtin_options_type: LSTMOptions,
// CHECK-NEXT:      builtin_options: {
// CHECK-NEXT:        fused_activation_function: RELU,
// CHECK-NEXT:        cell_clip: 1.0,
// CHECK-NEXT:        proj_clip: 2.0,
// CHECK-NEXT:        kernel_type: BASIC
// CHECK-NEXT:      },
// CHECK-NEXT:      intermediates: [ ]
// CHECK-NEXT:    } ],

  MatMul \
  Neg \
  Sum \
  Sub \
  Div \
  DivNoNan \
  Exp \
  Sqrt \
  SqrtGrad \
  Log1p

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

${generate} \
  --category=resource_variable \
  VarHandleOp \
  ReadVariableOp \
  AssignVariableOp \
  DestroyResourceOp

${generate} \
  --category=io \

def mnist_model():
  """Creates a MNIST model."""
  model = tf.keras.models.Sequential()
  model.add(tf.keras.layers.Flatten())
  model.add(tf.keras.layers.Dense(128, activation='relu'))
  model.add(tf.keras.layers.Dense(10, activation='softmax'))
  return model


class TestModule(tf.Module):

  def __init__(self):
    super(TestModule, self).__init__()
    self.model = mnist_model()

  // CHECK-NOT: tfrt_fallback_async.copy_if_small
  %0 = tfrt_fallback_async.executeop key(0) cost(1024) device("/job:localhost/replica:0/task:0/device:CPU:0") "tf.Relu"(%arg) {T = f32} : 1
  %1 = tfrt_fallback_async.executeop key(0) cost(1024) device("/job:localhost/replica:0/task:0/device:CPU:0") "tf.Relu"(%arg) {T = f32} : 1
  tfrt.return %0, %1 : !tfrt_fallback.tf_tensor, !tfrt_fallback.tf_tensor